Pattern recognition system

• US 4,441,205 A
• Filed: 05/18/1981
• Issued: 04/03/1984
• Est. Priority Date: 05/18/1981
• Status: Expired due to Term

• Current Assignee: Kulicke & Soffa Investments, Inc.
• Original Assignee: Kulicke and Soffa Industries Incorporated
• Inventors: Negin, Michael, Berkin, George M.
• Primary Examiner(s): Boudreau, Leo H.
• Application Number:
  US06/264,481
• Time in Patent Office: 1,051 Days
• Field of Search: 382/8, 382/14, 382/15, 382/34, 382/18, 382/48, 343/5 MM, 179/1 SB, 179/1 SD, 364/448, 364/447, 364/449, 364/456, 364/488-491, 356/373, 356/375, 358/103, 358/107
• US Class Current: 382/151
• CPC Class Codes: G01B 11/024 by means of diode-array scanning
• CPC Class Codes: G06V 10/758 Involving statistics of pixels or of feature values, e.g. histogram matching

Abstract

An improved pattern recognition system is disclosed which converts rows and columns of digitized video signal elements indicative of light intensity falling on a camera into integrated signatures, one for each of two orthogonal directions. Signatures corresponding to effectively rotated images may also be provided. Portions of the signatures having the most useful information contained therein are selected by comparison with subsets of the total signatures in a teach mode. In a run mode, the thus selected sub signatures are compared with snapshot signatures of a target object, such as integrated circuit dice, for precise location of corresponding portions of the target object.

Claims

1. A system for comparing groups of digitized data, comprising:

a. means for processing a first group to obtain a first signature representative of a first information portion of said first group of data;

b. means for selecting a set of subsignatures from said signature, said subsignatures representing selected respective portions of said first signature;

c. means for selecting one of the set of said subsignatures which has the maximum relative information content, by successive comparison of the members of said set of subsignatures to said group;

d. means for processing a second group to obtain a second signature representative of a corresponding first information portion of said second group; and

e. means for comparing said selected subsignature to respective portions of said second signature in order to determine a comparison error for each said portion of said second group, and for designating the respective portion of said second group with respect to which the lowest comparison error is determined, as being the portion of said second group to which said selected subsignature corresponds.

2. The system as described in claim 1, wherein said means for selecting the one of said subsignatures which has the maximum relative information content comprises:

means for comparing each of said subsignatures of said set with respective data portions of said first signature and determining the comparison yielding the minimum error of all of such comparisons performed with respect to each subsignature of said set; and

means for selecting the subsignature of said set which has the maximum minimum error resulting from said comparing.

3. The system as described in claim 1, further comprising means for generating a system error signal as a function of the location of said designated respective portion relative to the location of said selected sub signature.

4. The system as described in claim 3, further comprising means for transforming said system error signal into a displacement signal representative of the displacement of the video view of a second group relative to said first group.

5. The system as described in claim 1, wherein said means for selecting one of said subsignatures comprises means for selecting the subsignature of said set which has the highest comparison confidence level compared to other subsignatures of said set.

6. The system as described in claim 5, comprising error function means for comparing each sub signature of said set with respective portions of said signature to obtain a set of error functions.

7. The system as described in claim 6, wherein said error function means comprises guard band means to inhibit generation of error functions with respect to portions of said signature within a predetermined range of each said sub signature.

8. The system as described in claim 1, wherein each of said groups represents a video view.

9. The system as described in claim 1, further comprising at least one host machine for performing predetermined operations, and means for controlling at least one of said predetermined operations as a function of said designation.

10. A system for comparing groups of digitized data, comprising:

a. means for processing a first group to obtain a first signature representative of a first information portion of said first group of data;

b. means for selecting a set of subsignatures from said signature, said subsignatures representing selected respective portions of said first signature;

c. means for selecting one of said set of subsignatures;

d. means for processing a second group to obtain a second signature representative of a corresponding first information portion of said second group; and

e. means for comparing said selected subsignature to respective portions of said second signature in order to determine a comparison error for each said portion, and for designating the respective portion of said second group with respect to which the lowest comparison error is determined, as being the portion of said second group to which said selected subsignature corresponds;

wherein each of said groups of data represents a two dimensional image, said system further comprising means for obtaining first signatures with respect to each of two orthogonal directions of said first image, and for obtaining second signatures in each of two orthogonal directions of said second image, and for obtaining subsignatures of said first signatures.

11. The system as described in claim 10, comprising integration means for obtaining each of said signatures, by summing over image data for a group in a first orthogonal direction to obtain a first signature perpendicular thereto, and summing image data for said groups in the second orthogonal direction to obtain a signature perpendicular to said first signature.

12. The system of claim 11, further comprising rotating means for generating signatures representing an image effectively rotated relative to the first image data group.

13. The system of claim 12, wherein said rotating means comprises means for controlling said integration means to regularly step across said image, perpendicular to said direction of summing, at a rate determined by the desired degree of rotation.

14. The system of claim 10, wherein said digitized data corresponds to an image of a target object, and said system further comprises means for reperforming said steps (a)-(e) with respect to second groups of data corresponding to second image portions of said target object.

15. The system as described in claim 10, further comprising cross normalization means for normalizing said subsignatures as a function of said signature and for normalizing said signature as a function of respective said subsignatures.

16. A system for comparing a target set of digital data with a reference set of digital data, said sets representing patterns extending in two or more orthogonal directions, comprising:

means for processing the data of said reference set to obtain a reference signature comprising information representative of said reference set in a first of said orthogonal directions;

means for selecting a number of proposed subsignatures from said reference signature, each of said subsignatures containing less than all of the information of said reference signature;

means for selecting the subsignature of said set which has the highest relative information content;

means for proecessing said target set to obtain a target signature containing information representative of said target in said first direction;

means for comparing said subsignature having the highest relative information content with respective different subportions of said target signature and determining the subportion which provides the lowest comparison error; and

means for generating a signal representative of a spatial displacement in said first direction of said target set with respect to said reference set as a function of said determined subportion.

17. The system of claim 16 wherein said means for selecting the subsignature having the highest relative information content comprises means for comparing each of said numbers of proposed subsignatures with respective subportions of said reference signature and means for selecting that proposed subsignature which corresponds least closely to any one of said subportions of said reference signature.

18. The system of claim 17 wherein said means for comparison comprises means for summing the absolute values of the differences between respective elements of a proposed subsignature and elements of a subportion of said reference signature.

19. A system for comparing a target set of digital data with a reference set of digital data, said sets representing patterns extending in two or more orthogonal directions, comprising:

means for processing the data of said reference set to obtain a reference signature comprising information representative of said reference set in a first of said orthogonal directions;

means for selecting a number of proposed subsignatures from said reference signature, each of said subsignatures containing less than all of the information of said reference signature;

means for processing said target set to obtain a target signature containing information representative of said target in said first direction;

means for comparing said selected subsignatures with all other portions of said reference signature and selecting that subsignature which corresponds least closely with each other portion of said reference signature; and

means for generating a signal representative of a spatial displacement in said first direction of said target set with respect to said reference set as a function of said selected subsignature, wherein said patterns are video representations of physical objects and said data are gray scale values of video information in respective picture elements, and wherein said signatures are sums of the gray scale information contained in said picture elements, said sums being taken in both of said orthogonal directions to generate orthogonal signatures representative of the video image of a physical object.

20. The system of claim 19 wherein said second means for comparing comprises means for summing the absolute values of the differences between respective elements of each selected subsignature and of the corresponding element of the reference signature to which it is compared.

21. The system of claim 19 wherein means are additionally provided for generating an effectively rotated subsignature set from a reference signature corresponding to an effectively rotated portion of said signature.

22. The system of claim 21 wherein said sets of digital data representing patterns having two orthogonal directions comprise elements of data arrayed in each of said orthogonal directions, and are summed in each of said two orthogonal directions to generate said signatures, wherein said means for generating an effectively rotated subsignature set comprises means for iteratively summing over successions of a predetermined number of sequential ones of said data elements in said first orthogonal direction, said successions of predetermined numbers of data elements each being offset from one another by one element in the other of said orthogonal directions, whereby the degree of rotation effected is equal to the arc tangent of the inverse of said predetermined number.

23. A method for comparing like video images susceptible of relative rotation, comprising the steps of:

a. generating digital data representative of a first image;

b. generating first signatures representative of two orthogonal views of said first image;

c. generating rotated first signatures representative of two relatively rotated orthogonal views of said image;

d. selecting subsignature portions of each of said first signature and rotated first signatures, each of said subsignatures being selected by comparison of proposed subsignatures to plural portions of the corresponding signature, that one of said proposed subsignatures which corresponds least closely to any other portion of the corresponding signature being selected;

e. generating digital information representative of a second image;

f. generating second signatures representative of first and second orthogonal views of said second image and corresponding to said first signatures; and

g. comparing each of said selected subsignatures corresponding to said first signatures and said rotated first signatures respectively with each of said second signatures, and determining from said comparisons the rotation of said second view relative to said first view.

24. The method of claim 23 wherein said video images are divided into orthogonal rows and columns of individual pixels, each pixel having assigned thereto a digital value representative of the gray scale of said video image, said signatures comprising elements, said elements each being generated by summing over rows or columns of said pixels, wherein said first signatures are generated by summing along the orthogonal axes of said pixels and said rotated signatures are generated by summing at an angle to said axes.

25. The method of claim 24 wherein said comparing of said first signatures and rotated first signatures with each of said second signatures is performed by iteratively comparing said subsignatures with respective portions of said second signatures to determine which of said first signatures and rotated first signatures corresponds most closely with said second signatures, whereby the degree of rotation of said second view relative to said first view is determined.

26. The method of claim 25 wherein said comparison of said subsignatures with said second signatures is performed by summing the absolute value of the differences between respective elements of said subsignatures and of said second signatures.

27. Method for locating a predetermined point with respect to an image comprising an array of image elements, said array being N rows deep by M columns wide, comprising the steps of:

summing over each element of each row, to generate a first signature of N sums;

summing over each element of each column to generate a second signature of M sums,

iteratively comparing plural subsignatures, each comprising portions of said signatures, to other portions of said signatures to find that subsignature most different from all other portions of said signature, and

thereafter using the subsignature thus found as a reference for comparison with like portions of comparable images.

28. The method of claim 27 wherein said iterative comparison is performed by summing the absolute values of the differences between corresponding elements of said subsignatures and of the signature portion compared, and taking as most different that subsignature for which the minimum sum derived by comparison to a particular portion of a signature is greatest.

29. The method of claim 28 wherein normalization is performed on said compared portions of signatures and subsignatures prior to said summing step.

30. The method of claim 29 wherein said normalization is performed by cross-multiplying the average value of said elements of said subsignature by the elements of the portions of the signature compared, and the average value of elements of the portion of the signature compared by the elements of the subsignature, prior to said summing step.

31. In a method of comparing a predetermined subportion of a first image to a larger portion of a like second image, to establish the spatial position of said second image with respect to a reference position located on said first image, the improvement which comprises:

in a first teach mode, analyzing the total first image to determine that subportion most useful in said comparison, and storing said subportion; and

in a second run mode, using said stored subportion for establishing the spatial position of said second image;

wherein said analyzing step comprises processing said image to yield information-containing signatures representative of said image and iteratively comparing selected subportions of said first signature to corresponding other portions of said first signature, and selecting for storage that subportion which least closely corresponds to all portions of said first signature other than itself.

32. The method of claim 31 wherein said signatures comprise summed image density values taken along a line in said image, and said comparison with respect to a given sub-portion and a given other portion reflects the sum of the absolute values of the differences between corresponding elements of said given sub-portion and said given section of said signatures.

33. In a method of comparing plural stored reference sub-images of an object with a target object, to determine displacement and angular rotation of said target with respect to said stored images, the improvement which comprises:

comparing a first stored reference sub-image with a first portion of an image of said target to determine displacement and angular rotation and using the result determined to yield a starting point for comparison of a second stored reference sub-image with a second portion of an image of said target, wherein said stored reference sub-images are selected to maximize their information content by iteratively comparing coded representations of portions of the reference image with similarly coded other portions of the reference image, and selecting for storage that coded representation of a portion of the image which corresponds least closely to all the other coded portions.

34. Method for generating a subsignature useful in relating a target image to an object image, comprising the steps of:

dividing the object image into a plurality of pixels organized as rows and columns;

summing the video value of the image in each cell along each row and along each column, to obtain vertical and horizontal signatures each comprising a series of summed video values equal in length to the number of rows and of columns respectively;

selecting a plurality of subsignatures of each of said horizontal and vertical signatures, said subsignatures comprising lesser numbers of said summed video values than said signatures;

comparing each of said selected subsignatures with other portions of the corresponding signature of like length and determining for each of said subsignatures, the degree of greatest similarity of each said subsignature to another portion of the corresponding signature; and

selecting the subsignatures having the least similarity to any other portion of the corresponding signature as those to be used in relating said object image to a target image.

35. The method of claim 34 wherein said step of comparing a subsignature to a portion of a signature of like length comprises the step of summing the absolute values of the differences of the video values in the corresponding pixels of the subsignature and the portion of the signature, whereby a relatively large sum indicates a relatively low degree of similarity.

36. An apparatus for generating a subsignature useful in relating a target image to an object image, comprising:

means for dividing the object image into a plurality of pixels organized as rows and columns;

means for summing the video value of the image in each cell along each row and along each column, to obtain vertical and horizontal signatures each comprising a series of summed video values equal in length to the number of rows and of columns respectively;

means for selecting a plurality of subsignatures of each of said horizontal and vertical signatures, said subsignatures comprising lesser numbers of said summed video values than said signatures;

means for comparing each of said selected subsignatures with other portions of the corresponding signature of like length and determining for each of said subsignatures, the degree of greatest similarity of each said subsignature to any other portion of the corresponding signature; and

means for selecting the subsignatures having the least similarity to any other portion of the corresponding signature as those to be used in relating said object image to a target image.

37. The apparatus of claim 36 wherein said means for comparing a subsignature to a portion of a signature of like length comprises means for summing the absolute values of the differences of the video values in the corresponding pixels of the subsignature and the portion of the signature, whereby a relatively large sum indicates a relatively low degree of similarity.