DESCRIBING OBJECTS USING EDGE-PIXEL-FEATURE DESCRIPTORS

• US 20150363663A1
• Filed: 01/21/2013
• Published: 12/17/2015
• Est. Priority Date: 01/21/2013
• Status: Active Grant

• Current Assignee: Datalogic IP Tech S.r.l. (Datalogic S.p.A.)
• Original Assignee: Datalogic S.p.A.
• Inventors: TOMBARI, Federico, FRANCHI, Alessandro, DI STEFANO, Lugi
• Granted Patent: US 9,495,607 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06V 10/46 Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features colour feature extraction G06V10/56
• CPC Class Codes: G06V 10/48 by mapping characteristic values of the pattern into a parameter space, e.g. Hough transformation
• CPC Class Codes: G06V 10/50 by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis

Abstract

During a description technique (100), a local descriptor for an object (300) is generated (122) by computing a 2-dimensional histogram (600) of pairs of angles (514, 516) between pairs of line segments (510, 512) that are aligned with edge pixels associated with the object (300). The pairs of line segments (510, 512) belong to a subset of k neighboring or proximate line segments (310). Moreover, this 2D histogram (600) may represent the relative displacement and the relative orientations of the pairs of line segments (510, 512) in the subset as weights in bins or cells defined by angular quantization values, and the 2D histogram (600 may exclude lengths of the line segments. Subsequently, the generated 2D histogram (600) may be compared (210) to predefined sets of descriptors in a model library that are associated with a set of objects, and the object may be identified (212) as one of the set of objects based on a group of match scores determined in the comparisons.

Claims

1. A computer-system-implemented method for generating at least one descriptor for an object, the method comprising:

receiving extracted line segments aligned with edge pixels associated with the object in an image;

determining orientations for the line segments;

identifying a subset of the line segments, wherein the subset includes k of the line segments that are proximate to a given line segment in the line segments;

using the computer system, calculating pairs of angles representing relative displacement and relative orientation of pairs of line segments in the subset, wherein a given pair of angles for a given pair of line segments in the subset includes;

a first angle at a location on a first line segment in the given pair of line segments between the first line segment and a second line segment in the given pair of line segments, and a second angle at a location on the second line segment between the second line segment and the first line segment; and

generating the at least one descriptor for the object by computing a 2-dimensional (2D) histogram of the pairs of angles based on cells defined by angular quantization values.

2. The method of claim 1, wherein the method further comprises:

receiving the image that includes the object; and

extracting the line segments aligned with the edge pixels associated with the object.

3. The method of claim 1, wherein receiving the line segments includes accessing the line segments in a computer-readable memory.

4. The method of claim 1, wherein at least some of the line segments are not coupled to a remainder of the extracted line segments.

5. The method of claim 1, wherein the orientations are determined based on contrast polarity of the line segments.

6. The method of claim 1, wherein an orientation of the given line segment includes a direction aligned with the given line segment, and a sign computed by taking a cross product of a gradient at the location on the given line segment and the direction.

7. The method of claim 1, wherein the k line segments in the subset are nearest neighbors to the given line segment.

8. The method of claim 1, wherein, prior to determining the orientations, the method further comprises filtering the line segments based on lengths of the line segments and a threshold value.

9. The method of claim 1, wherein the location on the first line segment and the location on the second line segment include one of midpoints of the first line segment and the second line segment;

and points of the first line segment and the second line segment; and

points on the first line segment and the second line segment separated by a minimum distance between the first line segment and the second line segment.

10. The method of claim 1, wherein calculating the given pair of angles includes an operation other than calculating a dot product between the first line segment and the second line segment.

11. The method of claim 1, wherein calculating the first angle includes rotating the first line segment clockwise over a connecting line segment that couples the first line segment and the second line segment;

and

wherein calculating the second angle includes rotating the second line segment clockwise over the connecting line segment.

12. The method of claim 1, wherein, for the given pair of angles, the 2D histogram of the pairs of angles includes weights in neighboring, adjacent cells to a cell that includes the given pair of angles;

wherein the weights are determined using bilinear interpolation; and

wherein a sum of the weights equals a constant.

13. The method of claim 1, wherein the method further comprises storing the 2D histogram of the pairs of angles in a computer-readable memory that includes a set of descriptors of objects;

and

wherein, for a given object, the set of descriptors includes at least one descriptor.

14. The method of claim 1, wherein the method further comprises generating a set of descriptors for the object by repeating the operations in the method for multiple subsets of the line segments.

15. The method of claim 1, wherein the method further comprises generating a set of descriptors for the object by repeating the operations in the method for multiple images of the object from different vantage points.

16. The method of claim 1, wherein the method further comprises generating a set of descriptors for the object by repeating the operations in the method for different values of k.

17. The method of claim 16, wherein the method further comprises:

comparing the set of descriptors to a predefined set of descriptors associated with a second object, and

identifying the object as the second object based on a group of matches in the comparisons.

18. The method of claim 17, wherein the object is identified as the second object without using lengths of the line segments in the comparison.

19. The method of claim 1, wherein the method further comprises associating at least a portion of the image with an object category based on an object classifier.

20. The method of claim 1, wherein the method further comprises:

comparing the descriptor to predefined sets of descriptors that are associated with a set of objects; and

identifying the object as one of the set of objects based on a group of match scores determined in the comparisons.

21. The method of claim 20, wherein the comparing includes:

determining match scores between the descriptor and the predefined sets of descriptors based on Euclidean distances between the descriptor and the predefined sets of descriptors; and

selecting a subset of matches based on the match scores using one of a Generalized Hough Transformation and robust homography estimation.

22-36. -36. (canceled)