Discrete curve symmetry detection

0Associated
Cases 
0Associated
Defendants 
0Accused
Products 
1Forward
Citation 
0
Petitions 
1
Assignment
First Claim
1. A computer accessible memory medium that stores program instructions for detecting symmetries of discrete curves, wherein the program instructions are executable by a processor to perform:
 applying a first mapping operator to a first discrete curve, thereby generating a first mapped discrete curve, wherein the mapping operator operates to amplify features of the first discrete curve;
computing a correlation of the first mapped discrete curve with each of a plurality of rotationally shifted versions of a second mapped discrete curve, thereby generating a corresponding plurality of correlation values, wherein each correlation value indicates a respective degree of correlation between the first mapped discrete curve and a respective rotationally shifted version of the second mapped discrete curve;
determining a minimum period of the correlation of the first and second discrete curves based on the plurality of correlation values;
determining a symmetry group of the first and second discrete curves based on the minimum period; and
outputting the determined symmetry group.
1 Assignment
0 Petitions
Accused Products
Abstract
System and method for detecting symmetries of discrete curves. A mapping operator is applied to a first discrete curve to amplify its features, generating a first mapped discrete curve. A correlation of the first mapped discrete curve with each of a plurality of rotationally shifted versions of a second mapped discrete curve is computed, generating a corresponding plurality of correlation values. A minimum period of the two curves is determined based on the correlation values, and, based on the minimum period, a symmetry group (SG) of the two curves is determined and output. If the two curves are the same curve, the SG is the rotational SG of the discrete curve. If the second curve is a reflection of the first, the SG is the mutual reflection SG of the first. If the first and second curves are different curves, the SG is the mutual SG of the two curves.
16 Citations
View as Search Results
Identifying randomly distributed microparticles in images to sequence a polynucleotide  
Patent #
US 9,135,497 B2
Filed 01/27/2012

Current Assignee
National Instruments Corporation

Sponsoring Entity
National Instruments Corporation

Object image search using validated submodel poses  
Patent #
US 6,411,734 B1
Filed 12/16/1998

Current Assignee
Cognex Corporation

Sponsoring Entity
Cognex Corporation

Object image search using submodels  
Patent #
US 6,324,299 B1
Filed 04/03/1998

Current Assignee
Cognex Corporation

Sponsoring Entity
Cognex Corporation

Machine vision methods and system for boundary pointbased comparison of patterns and images  
Patent #
US 6,381,366 B1
Filed 12/18/1998

Current Assignee
Cognex Corporation

Sponsoring Entity
Cognex Corporation

Matching of discrete curves under affine transforms  
Patent #
US 20040037467A1
Filed 08/20/2002

Current Assignee
National Instruments Corporation

Sponsoring Entity
National Instruments Corporation

Rotational symmetry detection for configurations of discrete curves  
Patent #
US 20040247182A1
Filed 06/05/2003

Current Assignee
National Instruments Corporation

Sponsoring Entity
National Instruments Corporation

System and method for image pattern matching using a unified signal transform  
Patent #
US 20030007690A1
Filed 04/10/2001

Current Assignee
National Instruments Corporation

Sponsoring Entity
National Instruments Corporation

System and method for performing edge detection in an image  
Patent #
US 20030053696A1
Filed 06/28/2001

Current Assignee
National Instruments Corporation

Sponsoring Entity
National Instruments Corporation

Image pattern matching utilizing discrete curve matching with a mapping operator  
Patent #
US 20030198389A1
Filed 10/03/2002

Current Assignee
National Instruments Corporation

Sponsoring Entity
National Instruments Corporation

Method and apparatus for image registration using large deformation diffeomorphisms on a sphere  
Patent #
US 6,633,686 B1
Filed 09/20/2000

Current Assignee
Washington University In St Louis

Sponsoring Entity
Washington University In St Louis

Image processing method and apparatus, and storage medium  
Patent #
US 6,650,778 B1
Filed 01/20/2000

Current Assignee
Canon Kabushiki Kaisha

Sponsoring Entity
Canon Kabushiki Kaisha

Automated inspection of objects undergoing general affine transformation  
Patent #
US 6,421,458 B2
Filed 08/28/1998

Current Assignee
Cognex Corporation

Sponsoring Entity
Cognex Corporation

System and method for generating a low discrepancy curve in a region  
Patent #
US 20020140700A1
Filed 06/08/2001

Current Assignee
National Instruments Corporation

Sponsoring Entity
National Instruments Corporation

Apparatus and method for determining the location and orientation of a reference feature in an image  
Patent #
US 6,240,218 B1
Filed 03/14/1995

Current Assignee
Cognex Corporation

Sponsoring Entity
Cognex Corporation

Boundary tracking method and apparatus to find leads  
Patent #
US 6,035,066 A
Filed 06/02/1995

Current Assignee
Cognex Corporation

Sponsoring Entity
Cognex Corporation

Machine vision method and apparatus for determining the position of generally rectangular devices using boundary extracting features  
Patent #
US 5,933,523 A
Filed 03/18/1997

Current Assignee
Cognex Corporation

Sponsoring Entity
Cognex Corporation

58 Claims
 1. A computer accessible memory medium that stores program instructions for detecting symmetries of discrete curves, wherein the program instructions are executable by a processor to perform:
applying a first mapping operator to a first discrete curve, thereby generating a first mapped discrete curve, wherein the mapping operator operates to amplify features of the first discrete curve; computing a correlation of the first mapped discrete curve with each of a plurality of rotationally shifted versions of a second mapped discrete curve, thereby generating a corresponding plurality of correlation values, wherein each correlation value indicates a respective degree of correlation between the first mapped discrete curve and a respective rotationally shifted version of the second mapped discrete curve; determining a minimum period of the correlation of the first and second discrete curves based on the plurality of correlation values; determining a symmetry group of the first and second discrete curves based on the minimum period; and outputting the determined symmetry group.  View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 56, 57, 58)
 42. A method for detecting symmetries of discrete curves, the method comprising:
applying a first mapping operator to a first discrete curve, thereby generating a first mapped discrete curve, wherein the mapping operator operates to amplify features of the first discrete curve; computing a correlation of the first mapped discrete curve with each of a plurality of rotationally shifted versions of a second mapped discrete curve, thereby generating a corresponding plurality of correlation values, wherein each correlation value indicates a respective degree of correlation between the first mapped discrete curve and a respective rotationally shifted version of the second mapped discrete curve; determining a minimum period of the correlation of the first and second discrete curves based on the plurality of correlation values; determining a symmetry group of the first and second discrete curves based on the minimum period; and outputting the determined symmetry group.  View Dependent Claims (43, 44, 45, 46, 47, 48, 49, 50, 51, 52)
 53. A method for detecting reflection symmetries of discrete curves, the method comprising:
applying a mapping operator to a discrete curve, thereby generating a mapped discrete curve, wherein the mapping operator operates to amplify features of the discrete curve; computing a correlation of the mapped discrete curve with each of a plurality of rotationally shifted versions of a reflected image of the mapped discrete curve, thereby generating a corresponding plurality of correlation values, wherein each correlation value indicates a respective degree of correlation between the mapped discrete curve and a respective rotationally shifted version of the reflected image of the mapped discrete curve; determining a minimum period of the correlation based on the plurality of correlation values; determining a reflection symmetry group of the discrete curve based on the minimum period; and outputting the determined reflection symmetry group.
 54. A computer accessible memory medium that stores program instructions for detecting rotational symmetries of discrete curves, wherein the program instructions are executable by a processor to perform:
applying a mapping operator to a discrete curve, thereby generating a mapped discrete curve, wherein the mapping operator operates to amplify features of the discrete curve; computing a correlation of the mapped discrete curve with each of a plurality of rotationally shifted versions of the mapped discrete curve, thereby generating a corresponding plurality of correlation values, wherein each correlation value indicates a respective degree of correlation between the mapped discrete curve and a respective rotationally shifted version of the mapped discrete curve; determining a minimum period of the correlation of the discrete curve based on the plurality of correlation values; and determining a symmetry group of the discrete curve based on the minimum period; and outputting the determined rotational symmetry group.
 55. A computer accessible memory medium that stores program instructions for detecting reflection symmetries of a discrete curve, wherein the program instructions are executable by a processor to perform:
applying a mapping operator to a discrete curve, thereby generating a mapped discrete curve, wherein the mapping operator operates to amplify features of the discrete curve; computing a correlation of the mapped discrete curve with each of a plurality of rotationally shifted versions of a reflected image of the mapped discrete curve, thereby generating a corresponding plurality of correlation values, wherein each correlation value indicates a respective degree of correlation between the mapped discrete curve and a respective rotationally shifted version of the reflected image of the mapped discrete curve; determining a minimum period of the correlation of the discrete curve based on the plurality of correlation values; and determining a reflection symmetry group of the discrete curve based on the minimum period; and outputting the determined reflection symmetry group.
1 Specification
The present invention relates to geometric pattern matching in general, and more particularly to a system and method for detecting symmetries in and among discrete curves.
In many applications it is necessary or desired to determine the presence of an object of interest in a data set, such as a target image. For example, in many image processing applications it is desirable to find one or more matches of a template image in a larger target image. Exemplary machine vision applications include process monitoring, feedback control, and laboratory automation; image and video compression; and jitter compensation in video cameras, among others. Various characteristics may be used in classifying a location in the target image as a match, including luminance pattern information, color pattern information, and color information.
Additionally, the object of interest in the target data set or image may be transformed relative to the known object information, e.g., in the template data set or image. For example, the object of interest in the target image may be shifted, scaled, rotated, or may have other geometric or topological transformations.
Prior art pattern recognition systems have typically used a template matching technique wherein the stored image or pattern to be located is iteratively compared with various portions of a target image in which it is desired to locate the template.
Typically, the pattern matching algorithm involves comparing the template image, or a subset of sample pixels representing the template image, against locations in the target image on a horizontal pixel column basis and horizontal scan line basis. In other words, the sample pixels representing the template image are compared against a portion of the pixels in the target image, such as by using a 2D correlation, the sample pixels representing the template are then moved down or across a one pixel scan line or one pixel column in the target image, and the pattern matching algorithm is repeated, etc. Thus, the pattern matching algorithm generally involves comparing the template image pixels against all possible locations in the target image in an iterative fashion. The pattern matching may produce the location of the match in the image, the quality of match and possibly the orientation, size and/or scaling of the match.
The template is typically compared with portions of the target image by utilizing a correlation based pattern matching, i.e., using normalized two dimensional correlation (normalized 2D correlation). This 2D correlation is performed by placing the template over the respective portion of the image and performing a complete normalized 2D correlation between the pixels in the template and the pixels in the corresponding portion of the image, using values associated with the pixels, such as grayscale values. This correlation generally produces a correlation value that indicates the degree of correlation or match. For example, the correlation value may range between −1 and +1, wherein +1 indicates a complete match, 0 indicates no match, i.e., that the two images are uncorrelated, and −1 indicates that the two images are anticorrelated, i.e., a complete reversal of a match.
The normalized 2D correlation operation is based on a pointwise multiplication wherein the template is first conceptually placed over a portion of the image, the value associated with each point or pixel of the template is multiplied with the corresponding pixel value in the respective portion of the target image, and the result is summed over the entire template. Also, as noted above, the template image is generally compared with each possible portion of the target image in an iterative fashion. This approach is thus very computationally intensive.
Various optimizations or algorithms have been developed to provide a more efficient pattern matching technique. One prior art technique is to use selected samples or pixels from the template image, referred to as sample pixels, to represent the template image and hence to reduce the number of computations in the correlation.
Another prior art technique for performing pattern matching utilizes hue plane or color information, either alone or in combination with pattern matching. Utilizing color information can often be used to simplify a grayscale pattern matching problem, e.g., due to improved contrast or separation of an object from the background. Also, some applications may utilize color information alone, i.e., not in conjunction with pattern information, to locate target image matches, e.g., for cases when an application depends on the cumulative color information in a region and not how the colors are arranged within the region or the spatial orientation of the region.
In machine vision applications, color is a powerful descriptor that often simplifies object identification and extraction from a scene. Color characterization, location, and comparison is an important part of machine vision and is used in a large class of assembly and packaging inspection applications. Inspection involves verifying that the correct components are present in the correct locations. For example, color information may be used in inspecting printed circuit boards containing a variety of components; including diodes, resistors, integrated circuits, and capacitors. These components are usually placed on a circuit board using automatic equipment, and a machine vision system is useful to verify that all components have been placed in the appropriate positions.
As another example, color information is widely used in the automotive industry to verify the presence of correct components in automotive assemblies. Components in these assemblies are very often multicolored. For example, color characterization may be used to characterize and inspect fuses in junction boxes, i.e., to determine whether all fuses are present and in the correct locations. As another example, it is often necessary to match a fabric in one part of a multicolor automobile interior. A color characterization method may be used to determine which of several fabrics is being used.
Another prior art technique for performing pattern matching is referred to as geometric pattern matching, which may also be referred to as curve matching or shape matching. Geometric pattern matching generally refers to the detection and use of geometric features in an image, such as boundaries, edges, lines, etc., to locate geometrically defined objects in the image. The geometric features in an image may be reflected in various components of the image data, including, for example, luminance (grayscale intensity), hue (color), and/or saturation. Typically, geometric features are defined by boundaries where image data changes, e.g., where two differently colored regions abut. Geometric pattern matching techniques are often required to detect an object regardless of scaling, translation, and/or rotation of the object with respect to the template image. For further information on shape or geometric pattern matching, see “StateoftheArt in Shape Matching” by Remco C. Veltkamp and Michiel Hagedoorn (1999), and “A Survey of Shape Analysis Techniques” by Sven Loncaric, which are both incorporated herein by reference.
An issue that arises in many pattern matching applications is that the image objects being analyzed, e.g., the template images and/or the target images, may include various symmetries, e.g., rotational symmetries, where, for example, a first object is a rotated version of a second object, or where a configuration of equivalent objects includes rotational symmetries, and/or reflection symmetries, where a first object is a reflected version of a second object. Current methods for determining such symmetries are computationally intensive and error prone, in that correlations between rotated and/or reflected versions are often difficult to distinguish.
Therefore, improved systems and methods are desired for detecting symmetries in and between image objects, e.g., discrete curves. Additionally, improved methods are desired for preparing discrete curves for comparison, e.g., for a symmetry detection process, e.g., as part of a pattern matching application.
Various embodiments of a system and method for determining symmetries of discrete curves, such as rotational symmetries and/or reflection symmetries, are presented. In one embodiment, a mapping operator may be applied to a first discrete curve, thereby generating a corresponding first mapped discrete curve, where the mapping operator operates to amplify features of the discrete curve. In one embodiment, the first aid second mapped discrete curves may also be normalized, thereby generating normalized mapped discrete curves. In one embodiment, where a normalized discrete curve comprises a sequence of N points, D=(d_{1}, . . . , d_{N}), the mapping operator may be determined based on the normalized discrete curve by calculating w such that:
for k=0, . . . , N
w_{k+1}−w_{k}≅0 for k=0, . . . ,N
substantially holds, where w_{N+1}=W_{1}, and where δ_{ko }is the Kronecker delta.
Then, a correlation of the first mapped discrete curve with each of a plurality of rotationally shifted versions of a second mapped discrete curve may be computed, thereby generating a corresponding plurality of correlation values, where each correlation value indicates a respective degree of correlation between the first mapped discrete curve and a respective rotationally shifted version of the second mapped discrete curve, as is well known in the art. Note that as used herein, the term “correlation” refers to both equivalence and orientation. In other words, as is well known in the art, “similar” means that two objects are identical but for possible shift, scale, and rotation; “equivalent” means that two objects are identical but for possible shift and rotation, i.e., that they are in the same equivalence class, in mathematical terms; and “correlated” means that two objects are identical but for a possible shift, or, said another way, that the two objects are equivalent and oriented in the same way.
In one embodiment, the method may also include generating the second mapped discrete curve, e.g., by applying a second mapping operator to a received second discrete curve, thereby generating the second mapped discrete curve, wherein the second mapping operator operates to amplify features of the second discrete curve. For example, the second mapping operator may be computed based on the received second discrete curve, and applied to the second discrete curve to generate the second mapped discrete curve.
In a preferred embodiment, the first discrete curve and the second discrete curve comprise closed discrete curves, wherein the first discrete curve and the second discrete curve are interpreted as curves in a complex plane. Additionally, respective points of the first discrete curve and the second discrete curve are preferably ordered in a mathematically positive manner.
In a preferred embodiment, the first mapped discrete curve includes a first sequence of N points, D_{1}′=(d_{11}, . . . , d_{1N}), and the second mapped discrete curve includes a second sequence of N points, D_{2}′=(d_{21}, . . . , d_{2N}), and computing a correlation of the first mapped discrete curve with each of the plurality of rotationally shifted versions of the second mapped discrete curve includes computing:
for k=0, . . . , N;
where each s_{k }comprises a respective correlation value between the first mapped discrete curve and a respective rotationally shifted version of the second mapped discrete curve, and where the index k indicates the rotational shift.
Then, a minimum period of the two discrete curves may be determined based on the plurality of correlation values. The determination of the minimum period may be accomplished in a variety of ways. For example, in one embodiment, a plurality of maximum correlation values may be determined, where each of the plurality of maximum correlation values corresponds to a respective rotational shift between the discrete curves, and a relative rotational shift corresponding to successive maximum correlation values determined, where the relative rotational shift comprises the minimum period. In a similar embodiment, peak detection may be performed on the plurality of correlation values to determine two or more successive peaks, and the minimum period determined based on the determined two or more successive peaks. In one embodiment, the plurality of correlation values may be filtered prior to performing peak detection, thereby reducing peaks resulting from random effects. For example, a SavitzkyGolai filter of an order that avoids random effects may be used to filter the correlation data.
In another embodiment, determining a minimum period of the discrete curves based on the plurality of correlation values may include computing a power spectrum of the plurality of correlation values, and determining the minimum period based on the power spectrum. For example, as is well known in the art, the power spectrum may be computed via application of a Fourier transform to the data. The minimum period may be ascertained by determining a dominant peak in the power spectrum, e.g., by testing against a specified threshold.
In yet another embodiment, determining a minimum period of the discrete curves based on the plurality of correlation values may include determining two or more zero crossing points in the plurality of correlation values, and determining the minimum period based on the determined two or more zero crossing points. For example, in one embodiment, DC components from the plurality of correlation values may be removed to generate a modified set of correlation values, then one or more filters may be applied to the modified set of correlation values to generate filtered correlation values. The two or more zero crossing points may then be determined based on the filtered correlation values to estimate an underlying principal frequency of the filtered correlation values. Finally, the minimum period may be determined based on the estimated underlying principal frequency.
For example, if the two curves have a rotational symmetry, i.e., a period, of 90 degrees, then they also have a rotational symmetry or period of 180 degrees. The minimal period is thus 90 degrees
Then, a mutual symmetry group of the two discrete curves may be determined based on the minimum period. In one embodiment, determining the symmetry group based on the minimum period includes dividing one rotational cycle by the minimum period, thereby generating the symmetry group. In this example, the symmetry group comprises a rotational symmetry group. For example, if the minimum period is 90 degrees, then the two curves have a mutual rotational symmetry group of four, i.e., 360/90.
Finally, the indicated mutual symmetry group may be output, e.g., to a file, display device, process, such as a pattern matching process, an external system, etc.
Note that in some embodiments, the first and second discrete curves may be normalized at one or more stages of the process. For example, in one embodiment, the first discrete curve may be normalized prior to applying the mapping operator to the first discrete curve. Similarly, in one embodiment, the second discrete curve may be normalized prior to applying the second mapping operator to the second discrete curve. Similarly, in one embodiment, the first and second mapped discrete curves may be normalized prior to computing the correlation of the first mapped discrete curve with each of the plurality of rotationally shifted versions of the second mapped discrete curve. More generally, in various embodiments, various preprocessing steps may be applied to the first and second discrete curves prior to, or during, the determination of the mutual symmetry group for the two discrete curves, as described below.
According to various embodiments of the method described above, the determined symmetry group may be a rotational symmetry group or a reflection symmetry group. For example, in one embodiment, the second mapped discrete curve is a reflected version or image of a third mapped discrete curve, where, for example, the third mapped discrete curve is generated by applying a respective mapping operator to a received second discrete curve. In this case, the determined mutual symmetry group is a mutual reflection symmetry group of the first discrete curve and the second discrete curve.
Thus, in one embodiment, the method may include generating the second mapped discrete curve by applying a second mapping operator to the receive second discrete curve, thereby generating the third mapped discrete curve, and reflecting the third mapped discrete curve about a specified axis of reflection, thereby generating the second mapped discrete curve. generating the second mapped discrete curve by reflecting each point in the third mapped discrete curve about the specified axis, thereby generating respective points in the second mapped discrete curve. In one embodiment, reflecting each point in the third mapped discrete curve about the specified axis includes shifting and rotating a coordinate system in the complex plane such that the specified axis of rotation comprises a positive vertical axis through an origin of the complex plane, determining corresponding positions for each point in the third mapped discrete curve based on the shifted and rotated coordinate system, thereby generating corresponding transformed points, and computing a complex conjugate for of the transformed points, thereby generating corresponding reflected points, where the reflected points comprise the second mapped discrete curve.
Note that if the second mapped discrete curve is not a reflected version or image, then the determined symmetry group is a rotational symmetry group.
In an embodiment where the second mapped discrete curve comprises a reflected version of a third mapped discrete curve about a specified axis of reflection, determining the symmetry group based on the minimum period may include dividing one rotational cycle by the minimum period, thereby generating the symmetry group, wherein the symmetry group comprises a reflection symmetry group. Furthermore, in one embodiment, the method may include determining zero or more axes of reflection based on the specified axis of reflection and the minimum period.
For example, in one embodiment, determining zero or more axes of reflection based on the specified axis of reflection and the minimum period may include determining a rotational shift between an initial orientation of the second mapped discrete curve and an orientation associated with a first correlation value peak, where the determined rotational shift indicates a reflection axis offset, e.g., equal to half the rotational shift. An initial axis of reflection may be determined based on the specified axis of reflection and the reflection axis offset, and zero or more additional axes of reflection determined based on the initial axis of reflection and the minimum period.
In one embodiment, the specified axis has a first orientation angle, and determining the initial axis of reflection based on the specified axis of reflection and the reflection axis offset includes computing an orientation of the initial axis of reflection by adding the reflection axis offset to the first orientation angle. In this case, determining the zero or more additional axes of reflection based on the initial axis of reflection and the minimum period includes computing a respective zero or more orientations of the zero or more additional axes of reflection by adding successive multiples of the minimum period to the orientation of the initial axis. In other words, once the initial axis of reflection is determined using the reflection axis offset, any remaining axes of reflection may be determined by successively adding the minimum period until a full cycle has been covered.
For example, if the specified axis of reflection is at 30 degrees with respect to the xaxis, and the rotational shift between an initial orientation of the second mapped discrete curve and an orientation associated with a first correlation value peak is 20 degrees, then the reflection axis offset is equal to 10 degrees (20/2). Thus, the initial axis of reflection is simply 30 degrees plus the 10 degree offset, giving an orientation of 40 degrees. If the minimum period determined from the correlation values is equal to 120 degrees, then the additional axes of reflection are at 160 degrees and 280 degrees, respectively.
It should be noted that if a first discrete curve D_{1 }is equivalent to a second discrete curve D_{2}, then D_{2 }is equivalent to D_{1}. Moreover, from a purely mathematical standpoint, the property of generating maximal magnitudes of 1 in the sense of the correlation equation above is transitive, and so, classes of similar curves may be determined from a collection or set of received discrete curves.
Additionally, in some embodiments, the above techniques may be applied to a single curve to determine the curve'"'"'s symmetry groups under rotation and/or reflection. For example, a mapping operator may be applied to a discrete curve, thereby generating a mapped discrete curve, where the mapping operator operates to amplify features of the discrete curve.
A correlation of the mapped discrete curve with each of a plurality of rotationally shifted versions of the mapped discrete curve may be computed, thereby generating a corresponding plurality of correlation values, where each correlation value indicates a respective degree of correlation between the mapped discrete curve and a respective rotationally shifted version of the mapped discrete curve, as is well known in the art. As noted above, the term “correlation” refers to both equivalence and orientation, i.e., a high correlation value indicates that the two objects are equivalent and oriented in the same way.
As described above, in one embodiment, the rotated versions of the discrete curve may also be reflected. In other words, prior to computing the correlation values, the mapped discrete curve may be reflected about a specified axis, as described above, and the correlation values computed between the mapped discrete curve and the reflected version of the mapped discrete curve. In another embodiment, the discrete curve may be reflected prior to application of the mapping operator, where, for example, the mapping operator is applied as part of the correlation computation.
A minimum period of the discrete curve may be determined based on the plurality of correlation values, as described above in detail. Then, a symmetry group of the discrete curve may be determined based on the minimum period. As described above, the determination of the minimum period may be accomplished in a variety of ways, including, for example, determining a relative rotational shift corresponding to successive maximum correlation values, two or more successive peaks, via computation of a power spectrum of the plurality of correlation values, and/or by determination of zero crossing points, among others. As also described above, the symmetry group of the discrete curve may be a rotational symmetry group, or a reflection symmetry group, depending on whether the discrete curve was reflected prior to the correlation computation. As also noted above, in other embodiments, both approaches may be performed, resulting in a rotational symmetry group and a reflection symmetry group for the discrete curve. Finally, the indicated symmetry group may be output, e.g., to a file, display device, process, such as a pattern matching process, an external system, and so forth.
It should be further noted that the symmetry groups described above relate only to geometrical properties of the discrete curves (objects), and that in some applications, additional data may be associated with the discrete curves that may also come into play regarding object symmetries, such as, for example, color or gray scale information. For example, in a pattern matching application where an image is analyzed for symmetry, it may be the case that the above methods indicate that a first object in the scene is mutually symmetric (e.g., rotational or reflection) with respect to a second object (in a geometrical sense) in the scene. However, if the first object is blue and the second object is red, then clearly, no amount of rotation or reflection will make the objects equivalent with the same orientation. Thus, in some embodiments, the above determined symmetries may be considered to be necessary, but not sufficient, conditions for more general “image symmetries” or “object symmetries”, where further information besides geometrical data is used.
The techniques described herein are broadly applicable to any domains or fields of use where data sets, such as images or any other type of data, are compared and characterized with respect to symmetry. For example, applications contemplated include, but are not limited to, image processing, data mining, machine vision, e.g., related to robotics, automated manufacturing and quality control, etc., data analysis, and optical character recognition (OCR), among others.
Thus, in various embodiments, the method operates to detect symmetries in and/or among discrete curves, e.g., representing image objects.
A better understanding of the present invention can be obtained when the following detailed description of the preferred embodiment is considered in conjunction with the following drawings, in which:
While the invention is susceptible to various modifications and alternative forms specific embodiments are shown by way of example in the drawings and are herein described in detail. It should be understood, however, that drawings and detailed description thereto are not intended to limit the invention to the particular form disclosed. But on the contrary the invention is to cover all modifications, equivalents and alternatives falling within the spirit and scope of the present invention as defined by the appended claims.
The following patent applications are hereby incorporated by reference in their entirety as though fully and completely set forth herein:
U.S. patent application Ser. No. 10/263,560 titled “Pattern Matching System Utilizing Discrete Curve Matching with a Mapping Operator”, filed Oct. 3, 2002.
U.S. patent application Ser. No. 10/454,940 titled “Rotational Symmetry Detection for Configurations of Discrete Curves”, filed Jun. 2, 2003.
The following publications are hereby incorporated by reference in their entirety as though fully and completely set forth herein:
The National Instruments IMAQ™ IMAQ Vision Concepts Manual; and
“Efficient Matching Of Discrete Curves”, by Lothar Wenzel, National Instruments, Austin, Tex.
Terms
The following is a glossary of terms used in the present application:
Discrete Curve—a sequence of points that defines a simple, i.e., nonselfintersecting, curve, edge, or boundary in an image or other data. A discrete curve may be 2dimensional, 3dimensional, or of higher dimensionality, and may be open or closed (e.g., forming a loop). A discrete curve may be generated by performing edge or boundary detection on an image or other data set, and may be normalized with respect to number of points, distribution of points, energy, average position, and/or length of the discrete curve.
Mapping Operator—an operator that may be applied to a discrete curve, i.e., to the points in a discrete curve, to amplify or magnify features of the discrete curve, e.g., to facilitate rotational symmetry detection, pattern matching, and/or other geometric analyses.
Memory Medium—any of various types of memory devices or storage devices. The term “memory medium” is intended to include an installation medium, e.g., a CDROM, floppy disks 104, or tape device; a computer system memory or random access memory such as DRAM, DDR RAM, SRAM, EDO RAM, Rambus RAM, etc.; or a nonvolatile memory such as a magnetic media, e.g., a hard drive, or optical storage. The memory medium may comprise other types of memory as well, or combinations thereof. In addition, the memory medium may be located in a first computer in which the programs are executed, or may be located in a second different computer which connects to the first computer over a network, such as the Internet. In the latter instance, the second computer may provide program instructions to the first computer for execution. The term “memory medium” may include two or more memory mediums which may reside in different locations, e.g., in different computers that are connected over a network.
Carrier Medium—a memory medium as described above, as well as signals such as electrical, electromagnetic, or digital signals, conveyed via a communication medium such as a bus, network and/or a wireless link.
Programmable Hardware Element—includes various types of programmable hardware, reconfigurable hardware, programmable logic, or fieldprogrammable devices (FPDs), such as one or more FPGAs (Field Programmable Gate Arrays), or one or more PLDs (Programmable Logic Devices), such as one or more Simple PLDs (SPLDs) or one or more Complex PLDs (CPLDs), or other types of programmable hardware. A programmable hardware element may also be referred to as “reconfigurable logic”.
Medium—includes one or more of a memory medium, carrier medium, and/or programmable hardware element; encompasses various types of mediums that can either store program instructions/data structures or can be configured with a hardware configuration program.
Program—the term “program” is intended to have the full breadth of its ordinary meaning. The term “program” includes 1) a software program which may be stored in a memory and is executable by a processor or 2) a hardware configuration program useable for configuring a programmable hardware element.
Software Program—the term “software program” is intended to have the full breadth of its ordinary meaning, and includes any type of program instructions, code, script and/or data, or combinations thereof, that may be stored in a memory medium and executed by a processor. Exemplary software programs include programs written in textbased programming languages, such as C, C++, Pascal, Fortran, Cobol, Java, assembly language, etc.; graphical programs (programs written in graphical programming languages); assembly language programs; programs that have been compiled to machine language; scripts; and other types of executable software. A software program may comprise two or more software programs that interoperate in some manner.
Hardware Configuration Program—a program, e.g., a netlist or bit file, that can be used to program or configure a programmable hardware element.
Graphical Program—a program comprising a plurality of interconnected nodes or icons, wherein the plurality of interconnected nodes or icons visually indicate functionality of the program.
The following provides examples of various aspects of graphical programs. The following examples and discussion are not intended to limit the above definition of graphical program, but rather provide examples of what the term “graphical program” encompasses:
The nodes in a graphical program may be connected in one or more of a data flow, control flow, and/or execution flow format. The nodes may also be connected in a “signal flow” format, which is a subset of data flow.
Exemplary graphical program development environments which may be used to create graphical programs include LabVIEW, DasyLab, DiaDem and Matrixx/SystemBuild from National Instruments, Simulink from the MathWorks, VEE from Agilent, WiT from Coreco, Vision Program Manager from PPT Vision, SoftWIRE from Measurement Computing, Sanscript from Northwoods Software, Khoros from Khoral Research, SnapMaster from HEM Data, VisSim from Visual Solutions, ObjectBench by SES (Scientific and Engineering Software), and VisiDAQ from Advantech, among others.
The term “graphical program” includes models or block diagrams created in graphical modeling environments, wherein the model or block diagram comprises interconnected nodes or icons that visually indicate operation of the model or block diagram; exemplary graphical modeling environments include Simulink, SystemBuild, VisSim, Hypersignal Block Diagram, etc.
A graphical program may be represented in the memory of the computer system as data structures and/or program instructions. The graphical program, e.g., these data structures and/or program instructions, may be compiled or interpreted to produce machine language that accomplishes the desired method or process as shown in the graphical program.
Input data to a graphical program may be received from any of various sources, such as from a device, unit under test, a process being measured or controlled, another computer program, a database, or from a file. Also, a user may input data to a graphical program or virtual instrument using a graphical user interface, e.g., a front panel.
A graphical program may optionally have a GUI associated with the graphical program. In this case, the plurality of interconnected nodes are often referred to as the block diagram portion of the graphical program.
FIG. 3—Computer System
The computer system 102 may perform symmetry detection as part of a pattern characterization analysis of a template image and may use information determined in this analysis to determine whether a target image matches the template image and/or to locate regions of the target image which match the template image, with respect to pattern information. Images that are to be matched are preferably stored in the computer memory and/or received by the computer from an external device.
The computer system 102 preferably includes one or more software programs operable to perform the symmetry detection. The software programs may be stored in a memory medium of the computer system 102. The term “memory medium” is intended to include various types of memory, including an installation medium, e.g., a CDROM, or floppy disks 104, a computer system memory such as DRAM, SRAM, EDO RAM, Rambus RAM, etc., or a nonvolatile memory such as a magnetic medium, e.g., a hard drive, or optical storage. The memory medium may comprise other types of memory as well, or combinations thereof. In addition, the memory medium may be located in a first computer in which the programs are executed, or may be located in a second different computer which connects to the first computer over a network. In the latter instance, the second computer may provide the program instructions to the first computer for execution. Various embodiments further include receiving or storing instructions and/or data implemented in accordance with the foregoing description upon a carrier medium. Suitable carrier media include a memory medium as described above, as well as signals such as electrical, electromagnetic, or digital signals, conveyed via a communication medium such as networks and/or a wireless link.
Also, the computer system 102 may take various forms, including a personal computer system, mainframe computer system, workstation, network appliance, Internet appliance, personal digital assistant (PDA), television system or other device. In general, the term “computer system” can be broadly defined to encompass any device having a processor which executes instructions from a memory medium.
The software program(s) may be implemented in any of various ways, including procedurebased techniques, componentbased techniques, graphical programming techniques, and/or objectoriented techniques, among others. For example, the software program may be implemented using ActiveX controls, C++ objects, Java Beans, Microsoft Foundation Classes (MFC), or other technologies or methodologies, as desired. A CPU, such as the host CPU, executing code and data from the memory medium comprises a means for performing symmetry detection according to the methods or flowcharts described below.
In other embodiments, the techniques presented herein may be implemented in a medium configured to perform the described methods, such as, for example, a programmable hardware element, e.g., a Field Programmable Gate Array (FPGA).
FIG. 4—Machine Vision System
In the machine vision system of
FIG. 5—Image Acquisition System Block Diagram
As shown in
In this embodiment, the host computer system 102 also includes a video capture board 214 which is adapted for coupling to the video source 112. The video capture board 214 is preferably coupled to the peripheral bus 212. In addition to the video capture board 214, other peripheral devices (216 and 218) may be coupled to the peripheral bus 212, such as audio cards, modems, graphics cards, network cards, etc.
The video source 112 supplies the analog or digital video signals to the video capture board 214. The video capture board 214 transfers digitized video frames to the system memory 206 through peripheral bus 212 and bus bridge 204. In this embodiment, the video capture board 214 acquires the target image and transfers it to system memory 206. One or more regions of interest (ROI) may be specified in the target image which are desired to be analyzed for symmetries. In one embodiment, the resulting symmetry information may be used as part of a pattern matching process where the ROI is searched for regions having pattern information that matches the pattern information of a template image, or the entire target image may be searched.
The system memory 206 may store a template image. The system memory 206 may also receive and/or store one or more other images, such as selected regions of interest (ROIs) in the template image or another image, or acquired target images. The system memory 206 also preferably stores software according to the present invention which operates to analyze the pattern information of the template and target images, e.g., for detection of symmetries. The system memory 206 may store the pattern information of the template image for comparison to various regions in the target image during the symmetry detection process.
The term “image,” as used herein, may refer to any of various types of images. An image may be obtained from any of various sources, including a memory medium. An image may, for example, be obtained from an image file, such as a BMP, TIFF, AIPD, PNG, JPG, or GIF file, or a file formatted according to another image format. An image may also be obtained from other sources, including a hardware device, such as a camera, framegrabber, scanner, etc. An image may be a complex image, in which pixel values (positions) have a real part and an imaginary part.
It is noted that, in a symmetry detection application, the pattern information of the template image may be precalculated and stored in the computer, and the actual template image is then not required to be stored or used for subsequent symmetry detection determination/location operations with respective target images. Thus, when a target image is acquired, the software may compare the pattern information of the target image with the precomputed pattern information of the template image.
The present invention is preferably implemented in one or more software programs which are executable by a processor or CPU. The software program(s) of the present invention are preferably stored in a memory medium of a computer as described above.
Although many of the embodiments described herein relate to images and image processing, it is noted that the techniques described are broadly applicable to data sets and data processing. In other words, various embodiments of the invention may be used to perform discrete curve symmetry detection, where the discrete curves are determined from data as opposed to just images.
FIGS. 6A and 6B—Symmetry Groups
As is well known in the art, there are two types of finite groups in the Euclidean plane: the socalled cyclic groups S_{n }of order n and the dihedral groups D_{2n }of order n. All symmetry groups must coincide with one of them. The cyclic groups S_{n }refer to rotational symmetry groups, and the dihedral groups D_{2n }refer to reflection symmetry groups.
As used herein, the term “rotational symmetry group” refers to the number of degenerate rotational positions of an object or configuration, i.e., the number of equal angle rotations that can be applied to an object or configuration about its center that result in an equivalent state. For example, as is well known in the art, a square has a rotational symmetry group of four (90 degree rotations are degenerate, i.e., 90 degree symmetry), an equilateral triangle has a rotational symmetry group of three (120 degree symmetry), and so on. Thus, a regular polygon of n sides has a rotational symmetry group generated by rotations of 2π/n radians about the center of the polygon, or equivalently, 360/n degrees. Thus, at a resolution of one degree, a circle has a rotational symmetry group of 360.
Similarly, as used herein, the term “reflection symmetry group” is related to the number of axes about which an object may be reflected with no apparent effect. Note that an object'"'"'s reflection symmetry group number is actually twice the number of reflection axes for the object. Thus, an object with bilateral symmetry about only one axis has a reflection symmetry group of two.
Thus, in this example, the center square A_{0 }has an individual rotational symmetry group of four, the circle B_{0 }has an individual rotational symmetry group of 360 (where the rotational resolution is one degree), the peripheral squares C_{1}C_{6 }each have an individual rotational symmetry group of four, and the rectangles D_{1 }and D_{2 }each have an individual rotational symmetry group of two.
Now, each object may also be considered to have its own “configuration rotational symmetry group”, which is equal to the configuration rotational symmetry subgroup (of the configuration) to which the object belongs. For example, objects C_{1}C_{4 }each have a configuration rotational symmetry subgroup value of 4, since that subconfiguration'"'"'s rotational symmetry group is equal to each object'"'"'s rotational symmetry group. However, objects C_{5}C_{6 }each have a configuration rotational symmetry subgroup value of 2 (instead of 4), since the subconfiguration of those objects has a rotational symmetry group of 2 (180 degree symmetry).
As noted above, the rotational symmetry group of the configuration may also be considered. For example, as
Similarly, the configuration of
Each object may also be considered to have its own “configuration reflection symmetry group”, which is equal to the configuration reflection symmetry subgroup (of the configuration) to which the object belongs. For example, objects C_{1}C_{4 }each have a configuration reflection symmetry subgroup value of eight, since that subconfiguration'"'"'s reflection symmetry group is equal to each object'"'"'s reflection symmetry group. However, objects C_{5}C_{6 }each have a configuration reflection symmetry subgroup value of four (rather than eight), since subconfiguration of those objects has a reflection symmetry group of four (two axes of reflection).
As noted above, the reflection symmetry group of the configuration may also be considered. For example, as
FIG. 6B—Examples of Objects with Various Symmetry Groups
As
Object b), a closed curve “E” is shown with a single reflection axis (horizontal), and thus has reflection symmetry group D_{2}. As may be clearly seen, the object has no rotational symmetry group, since no rotation results in a degenerate position. Object c), a “B” curve, is similar to object b), with reflection symmetry group D_{2}, and no rotational symmetry group. Object d), a closed “N” curve, has no reflection group, but has a rotational symmetry group S_{2}, since it may be rotated 180 degrees to a degenerate position. Object e), a closed “S” curve, has no reflection symmetry group, but as a rotational symmetry group of S_{2}, since it may be rotated 180 degrees to a degenerate position. Object f), an asymmetric script “8”, has a reflection symmetry group D_{2}, since it may be reflected about a vertical axis, but has no rotational symmetry group. Object g), a square, has many reflection axes, with a reflection symmetry group of D_{8 }and a rotational symmetry group of S_{4}.
Note that if an object has two reflection axes, then the object has rotational symmetry as well, as shown in the example object a), the “H” image (two reflection axes, and 180 degree rotation symmetry).
FIGS. 7A7B—Detecting Symmetries Between Discrete Curves
As
In 704, a correlation of the first mapped discrete curve with each of a plurality of rotationally shifted versions of a second mapped discrete curve may be computed, thereby generating a corresponding plurality of correlation values, where each correlation value indicates a respective degree of correlation between the first mapped discrete curve and a respective rotationally shifted version of the second mapped discrete curve, as is well known in the art. Note that as used herein, the term “correlation” refers to both equivalence and orientation. In other words, as is well known in the art, “similar” means that two objects are identical but for possible shift, scale, and rotation; “equivalent” means that two objects are identical but for possible shift and rotation, i.e., that they are in the same equivalence class, in mathematical terms; and “correlated” means that two objects are identical but for a possible shift, or, said another way, that the two objects are equivalent and oriented in the same way.
In one embodiment, the method may also include generating the second mapped discrete curve, e.g., by applying a second mapping operator to a received second discrete curve, thereby generating the second mapped discrete curve, wherein the second mapping operator operates to amplify features of the second discrete curve. For example, as described below with reference to
In a preferred embodiment, the first discrete curve and the second discrete curve comprise closed discrete curves, wherein the first discrete curve and the second discrete curve are interpreted as curves in a complex plane. Additionally, respective points of the first discrete curve and the second discrete curve are preferably ordered in a mathematically positive manner.
In a preferred embodiment, the first mapped discrete curve includes a first sequence of N points, D_{1}′=(d_{11}, . . . , d_{1N}), and the second mapped discrete curve includes a second sequence of N points, D_{2}′=(d_{21}, . . . , d_{2N}), and computing a correlation of the first mapped discrete curve with each of the plurality of rotationally shifted versions of the second mapped discrete curve includes computing:
for k=0, . . . , N;
where each s_{k }comprises a respective correlation value between the first mapped discrete curve and a respective rotationally shifted version of the second mapped discrete curve, and where the index k indicates the rotational shift.
Then, a minimum period of the two discrete curves may be determined based on the plurality of correlation values, as indicated in 706. The determination of the minimum period may be accomplished in a variety of ways. For example, in one embodiment, a plurality of maximum correlation values may be determined, where each of the plurality of maximum correlation values corresponds to a respective rotational shift between the discrete curves, and a relative rotational shift corresponding to successive maximum correlation values determined, where the relative rotational shift comprises the minimum period. In a similar embodiment, peak detection may be performed on the plurality of correlation values to determine two or more successive peaks, and the minimum period determined based on the determined two or more successive peaks. In one embodiment, the plurality of correlation values may be filtered prior to performing peak detection, thereby reducing peaks resulting from random effects. For example, a SavitzkyGolai filter of an order that avoids random effects may be used to filter the correlation data.
In another embodiment, determining a minimum period of the discrete curves based on the plurality of correlation values may include computing a power spectrum of the plurality of correlation values, and determining the minimum period based on the power spectrum. For example, as is well known in the art, the power spectrum may be computed via application of a Fourier transform to the data. The minimum period may be ascertained by determining a dominant peak in the power spectrum, e.g., by testing against a specified threshold.
In yet another embodiment, determining a minimum period of the discrete curves based on the plurality of correlation values may include determining two or more zero crossing points in the plurality of correlation values, and determining the minimum period based on the determined two or more zero crossing points. For example, in one embodiment, DC components from the plurality of correlation values may be removed to generate a modified set of correlation values, then one or more filters may be applied to the modified set of correlation values to generate filtered correlation values. The two or more zero crossing points may then be determined based on the filtered correlation values to estimate an underlying principal frequency of the filtered correlation values. Finally, the minimum period may be determined based on the estimated underlying principal frequency.
For example, if the two curves have a rotational symmetry, i.e., a period, of 90 degrees, then they also have a rotational symmetry or period of 180 degrees. The minimal period is thus 90 degrees
Then, in 708, a mutual symmetry group of the two discrete curves may be determined based on the minimum period. In one embodiment, determining the symmetry group based on the minimum period includes dividing one rotational cycle by the minimum period, thereby generating the symmetry group. In this example, the symmetry group comprises a rotational symmetry group. For example, if the minimum period determined in 706 is 90 degrees, then the two curves have a mutual rotational symmetry group of four, i.e., 360/90.
Finally, in 710, the indicated mutual symmetry group may be output, e.g., to a file, display device, process, such as a pattern matching process, an external system, etc.
Note that in some embodiments, the first and second discrete curves may be normalized at one or more stages of the process. For example, in one embodiment, the first discrete curve may be normalized prior to applying the mapping operator to the first discrete curve. Similarly, in one embodiment, the second discrete curve may be normalized prior to applying the second mapping operator to the second discrete curve. Similarly, in one embodiment, the first and second mapped discrete curves may be normalized prior to computing the correlation of the first mapped discrete curve with each of the plurality of rotationally shifted versions of the second mapped discrete curve. More generally, in various embodiments, any or all of the preprocessing steps described below with reference to
According to various embodiments of the method of
Thus, in one embodiment, the method may include generating the second mapped discrete curve by applying a second mapping operator to the receive second discrete curve, thereby generating the third mapped discrete curve, and reflecting the third mapped discrete curve about a specified axis of reflection, thereby generating the second mapped discrete curve. generating the second mapped discrete curve by reflecting each point in the third mapped discrete curve about the specified axis, thereby generating respective points in the second mapped discrete curve. In one embodiment, reflecting each point in the third mapped discrete curve about the specified axis includes shifting and rotating a coordinate system in the complex plane such that the specified axis of rotation comprises a positive vertical axis through an origin of the complex plane, determining corresponding positions for each point in the third mapped discrete curve based on the shifted and rotated coordinate system, thereby generating corresponding transformed points, and computing a complex conjugate for of the transformed points, thereby generating corresponding reflected points, where the reflected points comprise the second mapped discrete curve.
Note that if the second mapped discrete curve is not a reflected version or image, then the determined symmetry group is a rotational symmetry group.
In an embodiment where the second mapped discrete curve comprises a reflected version of a third mapped discrete curve about a specified axis of reflection, determining the symmetry group based on the minimum period may include dividing one rotational cycle by the minimum period, thereby generating the symmetry group, wherein the symmetry group comprises a reflection symmetry group. Furthermore, in one embodiment, the method may include determining zero or more axes of reflection based on the specified axis of reflection and the minimum period.
For example, in one embodiment, determining zero or more axes of reflection based on the specified axis of reflection and the minimum period may include determining a rotational shift between an initial orientation of the second mapped discrete curve and an orientation associated with a first correlation value peak, where the determined rotational shift indicates a reflection axis offset, e.g., equal to half the rotational shift. An initial axis of reflection may be determined based on the specified axis of reflection and the reflection axis offset, and zero or more additional axes of reflection determined based on the initial axis of reflection and the minimum period.
In one embodiment, the specified axis has a first orientation angle, and determining the initial axis of reflection based on the specified axis of reflection and the reflection axis offset includes computing an orientation of the initial axis of reflection by adding the reflection axis offset to the first orientation angle. In this case, determining the zero or more additional axes of reflection based on the initial axis of reflection and the minimum period includes computing a respective zero or more orientations of the zero or more additional axes of reflection by adding successive multiples of the minimum period to the orientation of the initial axis. In other words, once the initial axis of reflection is determined using the reflection axis offset, any remaining axes of reflection may be determined by successively adding the minimum period until a full cycle has been covered.
For example, if the specified axis of reflection is at 30 degrees with respect to the xaxis, and the rotational shift between an initial orientation of the second mapped discrete curve and an orientation associated with a first correlation value peak is 20 degrees, then the reflection axis offset is equal to 10 degrees (20/2). Thus, the initial axis of reflection is simply 30 degrees plus the 10 degree offset, giving an orientation of 40 degrees. If the minimum period determined from the correlation values is equal to 120 degrees, then the additional axes of reflection are at 160 degrees and 280 degrees, respectively.
Thus, various embodiments of the method of
As
Note that if a reflection symmetry group is to be detected, i.e., rather than a rotational symmetry group, the method may include reflecting the second discrete curve (or the first discrete curve) about a specified axis, as described above with reference to
In 713, a determination may be made as to whether the first number of points is approximately equal to the second number of points. For example, if the number of points included in the first discrete curve is equal to the number of points included in the second discrete curve within a few percent, e.g., within approximately 1 to 3 percent, then the number of points in the two curves may be considered to be approximately equal, i.e., the first number of points and the second number of points may be considered to be close in value. If the first number of points and the second number of points are determined to not be close in value, then the method may terminate, as indicated in 720.
If, in 713, the first number of points and the second number of points are determined to be close in value, then in 714, respective symmetry groups may be determined for each discrete curve. In a preferred embodiment, the respective symmetry groups may be determined in accordance with the method of
In a preferred embodiment, the points of the first discrete curve and the second discrete curve are substantially uniformly distributed with respect to arclength. In other words, the points of the discrete curves are preferably uniformly spaced at substantially the same increment.
In 715, a determination may be made as to whether the determined symmetry groups for the two discrete curves are equal. If the symmetry groups are not equal, then the method may terminate, as indicated in 720.
If in 715, the symmetry groups for the two discrete curves are found to be equal, then in 717, a determination may be made as to whether the number of points in each discrete curve are equal, and if not, then in 718, the first discrete curve and the second discrete curve may be uniformly resampled such that the number of points in each discrete curve are equal.
In 706, a correlation of the first discrete curve with each of a plurality of rotationally shifted versions of the second discrete curve may be computed, thereby generating a corresponding plurality of correlation values, where each correlation value indicates a respective degree of correlation between the first discrete curve and a respective rotationally shifted version of the second discrete curve, as described above in 706 with reference to
It should be noted that in some embodiments, one or both of the two discrete curves may be normalized prior to computing the correlation values. For example, as described below in more detail with reference to 804 of
Then, in 719, a determination may be made as to whether the first discrete curve and the second discrete curve are substantially equivalent based on the plurality of correlation values. For example, in one embodiment, the correlation values may range from zero, indicating no correlation, to one, indicating complete correlation, i.e., curve equivalence with identical orientations. Thus, the first and second discrete curves may be considered substantially equivalent if maximum correlation values approaching one are determined.
If in 719 the two discrete curves are determined not to be substantially equivalent, then the method may terminate, as indicated in 720. If, however, the two discrete curves are found to be substantially equivalent, then in 708, a minimum period of the first discrete curve with respect to the second discrete curve based on the plurality of correlation values may be determined. Said another way, if the maximal magnitude of s_{k }calculated above is close to 1, D_{1 }and D_{2 }are substantially equivalent, and so the smallest absolute phase shift that realizes this maximum may be determined, i.e., the minimum period.
In 708, a mutual symmetry group of the first discrete curve and the second discrete curve may be determined, and where the first discrete curve and the second discrete curve are mutually symmetric under the indicated mutual symmetry group, as described above with reference to
Additionally, in some embodiments, the above techniques may be applied to a single curve to determine the curve'"'"'s symmetry groups under rotation and/or reflection, as described in more detail below with reference to
Finally, in 710, the determined mutual symmetry group of the first discrete curve and the second discrete curve may be output, e.g., for storage, transmission to an external system, and/or for use by another process, such as a pattern matching process, among others.
As mentioned above, in one embodiment, the techniques described above may be used to determine both rotational and reflection symmetry groups, e.g., as part of a pattern matching application. For example, in one embodiment, discrete curve matching between a first discrete curve and a second discrete curve using symmetry detection may be performed as follows:
A mutual rotational symmetry group for the first discrete curve and the second discrete curve may be determined, e.g., using an embodiment of the method of
In one embodiment, determining the mutual rotational symmetry group for the first discrete curve and the second discrete curve may include computing a plurality of correlation values between the first discrete curve and the second discrete curve. If a maximum correlation value of the plurality of correlation values indicates that the discrete curves are substantially equivalent, results indicating that the discrete curves are substantially equivalent may be output.
Similarly, in another embodiment, determining a mutual reflection symmetry group for the first discrete curve and the second discrete curve may include computing a plurality of correlation values between the first discrete curve and a reflected version of the second discrete curve. If a maximum correlation value of the plurality of correlation values indicates that the first discrete curve and the reflected version of the second discrete curve are substantially equivalent, results indicating that the first discrete curve and the reflected version of the second discrete curve are substantially equivalent may be output.
It should be noted that if a first discrete curve D_{1 }is equivalent to a second discrete curve D_{2}, then D_{2 }is equivalent to D_{1}. Moreover, from a purely mathematical standpoint, the property of generating maximal magnitudes of 1 in the sense of equation (1) is transitive, and so, classes of similar curves may be determined from a collection or set of received discrete curves. This aspect may be used in relation to configurations of discrete curves, as described below.
FIG. 8A—PreProcessing for Detecting Symmetry of a Discrete Curve
In some embodiments of the present invention, one or more preprocessing operations may be performed prior to, or as part of, the methods described herein, e.g., the methods of
In 802, the discrete curve may be received, where the discrete curve is a closed discrete curve, and where the discrete curve may be interpreted as a curve in a complex plane. In other words, each point coordinate in the curve is a complex number, and the sequence of points makes a simple closed loop.
Note that the discrete curve may be received from any source, including, for example, from memory, from another system or process, such as an image processing system, and so forth. In one embodiment, the rotational symmetry detection process may include image processing functions and/or other preprocessing operations.
For example, in one embodiment, the method may include acquiring a data set, and determining the discrete curve from the data set, where the discrete curve corresponds to a respective object in the data set. In one embodiment, the discrete curve may correspond to an object of interest in the data set. The data set may be any type of data set, although in most applications, the data set comprises image data. Determining the discrete curve from the data set may include performing edge detection, also referred to as boundary detection, on the data set to determine the discrete curve, and in some embodiments, may include applying a filter to smooth the discrete curve.
In one embodiment, the discrete curve may be preprocessed (prior to determining the mapping operator in 806 below), as indicated in 804. For example, in one embodiment, preprocessing may include normalizing the discrete curve, and computing the center of mass of the normalized discrete curve.
Discrete curves may be normalized in a variety of ways. For example, normalizing the discrete curve may include normalizing the number of points in the discrete curve, where, for example, the number of points is proportional to arclength (i.e., the perimeter) for the discrete curve; normalizing distribution of points in the discrete curve, e.g., a uniform distribution along the length of the curve; normalizing energy of the discrete curve, for example, by computing the center of mass of the discrete curve, and normalizing the distance from each point in the discrete curve to the center of mass such that the sum of the squares of the distances is equal to a specified value; normalizing average position of the discrete curve, e.g., such that the center of mass of the points is at the origin; and/or normalizing length of the discrete curve, e.g., where the total length of the curve is equal to one. Other normalization schemes are also contemplated.
Then, in 806, the mapping operator (used above in step 704 of the method of
In one embodiment, where the normalized discrete curve comprises a sequence of N points, D=(d_{1}, . . . , d_{N}), determining the mapping operator based on the normalized discrete curve may include calculating w such that:
substantially holds, where w_{N+1}=w_{1}, and where δ_{ko }is the Kronecker delta. In other words, equation (2) may be solved for w, e.g., in a least squares manner, such that w amplifies features, e.g., distinguishing features, of the curve. This type of amplification of features may be accomplished by distorting the curve to increase the range of correlation values between curves, or between a curve and rotated versions of itself, thereby making correlations more apparent.
FIGS. 8B and 8C—Correlation Effect of Mapping Operator
The top right image in the
Thus, application of a suitable mapping operator prior to correlation computations may substantially improve results of such computations.
FIG. 9—Detecting Symmetries of a Discrete Curve
The general approach described above with reference to
As
In 904, a correlation of the mapped discrete curve with each of a plurality of rotationally shifted versions of the mapped discrete curve may be computed, thereby generating a corresponding plurality of correlation values, where each correlation value indicates a respective degree of correlation between the mapped discrete curve and a respective rotationally shifted version of the mapped discrete curve, as is well known in the art. As noted above, the term “correlation” refers to both equivalence and orientation, i.e., a high correlation value indicates that the two objects are equivalent and oriented in the same way. For example, in a preferred embodiment, the correlations may be computed thusly:
for k=0, . . . , N,
where each s_{k }comprises a respective correlation value between the mapped discrete curve and a respective rotationally shifted version of the mapped discrete curve, where the index k indicates the rotational shift. Note that in equation (3), the application of the mapping operator (902) is performed in the correlation operation (904). In another embodiment, the mapping operator may be applied separately, and thus, the correlation equation may not include the mapping operator explicitly.
Note that in equation (3), if w_{n }is interpreted as the mapping operator, then, as shown, the mapping operator is applied only to one of the curves (e.g., to the original discrete curve or to the rotated version of the discrete curve). However, in an alternate interpretation, w_{n }may be interpreted as the square of the mapping operator, where the mapping operator is applied to both the discrete curve and the rotated versions of the discrete curve. In other words, in one embodiment, the mapping operator may be the square root of w_{n}, i.e., w_{n}^{1/2}. In this case, both the discrete curve and the rotated version of the discrete curve may be mapped respectively to a mapped discrete curve and a rotated version of the mapped discrete curve, and the two mapped curves correlated. It should be noted that this alternate approach is mathematically equivalent to mapping only one of the discrete curves using w_{n}. However, in this alternate interpretation, the symmetric application of the mapping operator (w_{n}^{1/2}) may be interpreted as mapping both discrete curves into forms that enhance differences between the curves.
Note than in an embodiment where the mapped discrete curve is normalized, computing the correlation of the mapped discrete curve with a plurality of rotationally shifted versions of the mapped discrete curve preferably includes computing the correlation of the normalized mapped discrete curve with a plurality of rotationally shifted versions of the normalized mapped discrete curve.
It should be noted that in equation (3), the computation of the correlation values and the application of the mapping operator are performed as part of a single calculation, and so in one embodiment, the operations of 904 and 906 may be combined into a single operation. In another embodiment, the mapping operator may be applied to the curve or curves as a separate step, in which case the mapping operator may not explicitly appear in equation (3).
As described above with reference to
In 906, a minimum period of the discrete curve may be determined based on the plurality of correlation values, as described above in detail.
Then, a symmetry group of the discrete curve may be determined based on the minimum period, as indicated in 908. As described in detail in 708 above with reference to
Finally, in 910, the indicated symmetry group may be output, e.g., to a file, display device, process, such as a pattern matching process, an external system, and so forth.
It should be further noted that the symmetry groups described above relate only to geometrical properties of the discrete curves (objects), and that in some applications, additional data may be associated with the discrete curves that may also come into play regarding object symmetries, such as, for example, color or gray scale information. For example, in a pattern matching application where an image is analyzed for symmetry, it may be the case that the above methods indicate that a first object in the scene is mutually symmetric (e.g., rotational or reflection) with respect to a second object (in a geometrical sense) in the scene. However, if the first object is blue and the second object is red, then clearly, no amount of rotation or reflection will make the objects equivalent with the same orientation. Thus, in some embodiments, the above determined symmetries may be considered to be necessary, but not sufficient, conditions for more general “image symmetries” or “object symmetries”, where further information besides geometrical data is used.
Configurations of Discrete Curves
As noted above with reference to
The configuration rotational symmetry group of a configuration relates to a minimal phase angle through which the entire configuration may be rotated that results in a degenerate or indistinguishable state of the configuration. In other words, after the rotation, the configuration is indistinguishable from its original state. Note that this configurationbased view of rotational symmetry constrains the individual rotational symmetry group of each object in the configuration to be consonant with those of the other objects in the configuration. For example, referring back to the configuration of
Additionally, as also described above with reference to
FIG. 10—Configuration Rotational Symmetry SubGroups in a Configuration of Equivalent Objects
As
FIG. 11—Determination of Configuration Rotational Symmetry SubGroups in a Configuration of Equivalent Objects
As
In 1104, the configuration may be normalized. In one embodiment, normalizing the configuration may include normalizing each of the plurality of discrete curves, and computing a respective discrete curve center of mass (average position) for each of the plurality of discrete curves. A configuration center of mass, i.e., a center of mass of the plurality of discrete curves, may also be computed, where distances between the configuration center of mass and each respective discrete curve center of mass are constants, and where each discrete curve center of mass comprises a respective coordinate in the complex plane. In other words, the configuration may be considered or interpreted to be in the complex plane, and so each discrete curve center of mass may have a complex coordinate in that plane. In a preferred embodiment, the configuration center of mass may be normalized to the origin.
Each discrete curve center of mass may then be normalized with respect to the configuration center of mass. In one embodiment, normalizing each discrete curve center of mass with respect to the configuration center of mass may include setting the distances between the configuration center of mass and each discrete curve center of mass to a specified value, and normalizing each discrete curve center of mass such that a first discrete curve center of mass may be located at the specified value on the positive real axis of the complex plane. For example, in a preferred embodiment, the specified value may be one, where the configuration center of mass is at the origin of the complex plane, and where each discrete curve center of mass is located on a unit circle centered at the configuration center of mass. In other words, the position of each discrete curve center of mass may be normalized so that it lies on a unit circle centered at the configuration center of mass, e.g., at the origin of the complex plane, as illustrated in the example configuration of
As noted above, normalizing each discrete curve may include normalizing one or more of: number of points in the discrete curve, e.g., such that the number of points is proportional to arclength for the discrete curve; distribution of points in the discrete curve, e.g., to a uniform distribution; energy of the discrete curve, e.g., by computing the center of mass of the discrete curve, and normalizing the distance from each point in the discrete curve to the center of mass such that the sum of the squares of the distances is equal to a specified value; average position of the discrete curve, e.g., to lie on the unit circle, as described above; and length of the discrete curve, e.g., to one. Other normalization schemes are also contemplated.
In 1106, a plurality of respective orientations between respective pairs of the discrete curves may be determined. For example, in one embodiment, the method described above with reference to

 for k=0, . . . , N,
where each s_{k }comprises a respective correlation value between the first discrete curve and a respective rotationally shifted version of the second discrete curve. Then, a minimum period of the two discrete curves may be determined based on the plurality of correlation values, as described above in 708 with reference to
In 1108, one or more finite subgroups, i.e., subsets, of the plurality of discrete curves may be determined based on the configuration and the plurality of respective orientations, where the one or more finite subgroups correspond to a respective one or more rotational symmetry subgroups of the configuration. For example, each finite subgroup may be characterized by each object or discrete curve in the subgroup having a common rotational symmetry group, and the subconfiguration of the objects or discrete curves in that subgroup having the same rotational symmetry group. In other words, the particular relative phase angles between elements in the finite subgroup coincide with the subconfiguration rotational symmetry angles. Said another way, in corresponding to a respective rotational symmetry subgroup of the configuration, the discrete curves of each finite subgroup may have relative orientations in accordance with the respective rotational symmetry subgroup.
For example, referring back to
In one embodiment, the method may also include removing each finite subgroup that is a subset of another finite subgroup, resulting in remaining finite subgroups. In other words, any harmonic rotational relationships that exist between the various determined finite subgroups may be used to remove finite subgroups of the configuration that are, in a sense, redundant, in that the configuration rotational symmetry group of the subset finite group is included in the configuration rotational symmetry group of the superset finite group.
Finally, in 1110, the one or more rotational symmetry subgroups of the configuration may be output, e.g., for transmission to an external system or process, to a display device, to a file, etc. In an embodiment where subset finite subgroups are removed, leaving remaining finite subgroups, the rotational symmetry subgroups of the configuration that correspond to the remaining finite subgroups may be output.
FIG. 12—Characterizing Rotational Symmetries of a Configuration of Equivalent Objects
As
In 1204, the configuration may be normalized, e.g., by computing a configuration center of mass, i.e., a center of mass of the plurality of discrete curves, and normalizing the configuration center of mass to a specified position. For example, in a preferred embodiment, the configuration may be interpreted as lying in the complex plane, and the configuration center of mass normalized to the origin of the complex plane.
In 1206, each discrete curve in the configuration may be classified with respect to shape, e.g., an equivalence class for each discrete curve may be determined, as described above in the methods of
In 1208, the center of mass of each discrete curve may be determined, as described above, and as is well known in the art.
In 1210, one or more subsets of the plurality of discrete curves may be determined based on each discrete curve'"'"'s classification (equivalence class) and distance from the configuration center of mass, where each subset includes one or more discrete curves of the same equivalence class at a respective distance from the configuration center of mass. In one embodiment, each respective distance from the configuration center of mass may be termed a “layer” of the configuration, where the configuration center of mass defines layer 0, the nearest distance defines layer 1, and so on.
Then, in 1212, one or more configuration rotational symmetry groups may be determined for each of the one or more subsets based on each discrete curve'"'"'s rotational symmetry group, and relative orientations of the discrete curves in the subset. For example, for each subset, as described above in 1108 with respect to
As also described above, in one embodiment, the method may also include removing each finite subgroup that is a subset of another finite subgroup, resulting in remaining finite subgroups.
Finally, in 1214, configuration characterization information generated in the above steps may be output, e.g., to another process, system, file, etc., as described above. For example, in one embodiment, one or more of: the determined one or more configuration rotational symmetry groups for each of the one or more subsets, the object or discrete curve classifications, the configuration layers of each object or discrete curve, and the respective rotational symmetry group for each discrete curve, among others, may be output.
FIGS. 13A and 13B—Example Results
As
The type of characterization information presented in
FIG. 14—Determining an Underlying Rotational Symmetry Group of a Configuration of Discrete Curves
As noted above, in various embodiments, some of the steps described may be performed concurrently, in a different order than shown, or may be omitted. Additional steps may also be performed as desired.
In 1402, configuration characterization information may be received corresponding to a configuration of discrete curves, also referred to as objects. In one embodiment, the configuration characterization information preferably includes the configuration rotational symmetry group of each object class per layer in the configuration. In another embodiment, the configuration characterization information may also include one more of the rotational symmetry group of each object class represented in the configuration and the layers of the configuration. One example of such configuration characterization information is provided in
In one embodiment, the configuration characterization information may be determined using one or more of the techniques described above with reference to
In 1404, a determination may be made as to whether an object in layer 0 of the configuration has a configuration rotational symmetry group of one. In other words, the method may determine if an object at the center of the configuration has a configuration rotational symmetry group of one, meaning that the object has no rotation symmetry, and thus that the configuration has no rotation symmetry. If a layer 0 object does have a configuration rotational symmetry group of one, then the method may terminate, as indicated in 720.
If no layer 0 object of the configuration has a configuration rotational symmetry group of one, then in 1404, a greatest common divisor (GCD) of the configuration rotational symmetry groups of the configuration, also referred to as a greatest common factor, may be determined. This value represents the underlying rotational symmetry group of the entire configuration. Note that if the GCD is equal to one, then the configuration as a whole has no rotational symmetries.
Finally, in 1406, the determined underlying rotational symmetry group of the configuration may be output, e.g., to file, to a display device, to another system or process, etc. In one embodiment, the results of the symmetry analysis may be presented in a graphical user interface, an example of which is shown in
FIG. 15—A Symmetry Detection GUI
In a preferred embodiment, the GUI provides a user interface to a program that determines the edges of the template image, e.g., the crosslike object shown, and based on the determined edges (discrete curves) a potential symmetry group of the image template is identified. In the embodiment of
In the example shown in
In an ideal world all these crosses would be 100% identical. However, as may be seen, there are minor differences between them. These differences may be quantified by comparing the original template image with the rotated versions using classical normalized crosscorrelation. Note that in a preferred embodiment, the crosscorrelation is based on image content, and not solely on detected edges of the image. In other words, in addition to the edges of the image, information such as color or gray scale pixel data may also be used.
The correlation process may generate a plurality of numbers indicating the degree to which each pair of images matches. For example, in one embodiment, the correlation process may generate 16 real numbers that should be close to 1 (assuming at least near matches among the image pairs. Actually, in a preferred embodiment, the number of unique correlation values generated may be somewhat less. More specifically, for n objects (images), n(n−1)/2 correlation values may be computed. Thus, in the case of the four images of
In the example shown in
Thus, in a preferred embodiment, the GUI provides a user interface for a program that generates the symmetry group of a given template and computes a quality measure representing the degree of similarity between the original template or configuration and rotated versions generated in accordance with the detected symmetry group of the configuration. Additionally, the GUI of
The techniques described herein are broadly applicable to any domains or fields of use where data sets, such as images or any other type of data, are compared and characterized with respect to symmetry. For example, applications contemplated include, but are not limited to, image processing, data mining, machine vision, e.g., related to robotics, automated manufacturing and quality control, etc., data analysis, and optical character recognition (OCR), among others.
It should be noted that in other embodiments, the methods described above regarding the determination of reflection symmetries of and between discrete curves may be applied to configurations of discrete curves, similar to the above described extension of rotational symmetries to configurations of discrete curves.
Although the embodiments above have been described in considerable detail, numerous variations and modifications will become apparent to those skilled in the art once the above disclosure is fully appreciated. It is intended that the following claims be interpreted to embrace all such variations and modifications.