System and method for identifying scale invariant features of object outlines on images

• US 9,135,522 B2
• Filed: 02/27/2012
• Issued: 09/15/2015
• Est. Priority Date: 02/27/2012
• Status: Active Grant

• Current Assignee: Aselsan Elektronik Sanayi Ve Ticaret Anonim Sirketi
• Original Assignee: Aselsan Elektronik Sanayi Ve Ticaret Anonim Sirketi
• Inventors: Akagunduz, Erdem
• Primary Examiner(s): MARIAM, DANIEL G
• Application Number:
  US14/240,316
• Publication Number:
  US 20140212048A1
• Time in Patent Office: 1,296 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 18/22 Matching criteria, e.g. proximity measures
• CPC Class Codes: G06T 9/20 Contour coding, e.g. using detection of edges
• 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

Abstract

The present invention relates to a system and method for identifying scale invariant features of image outlines. The method comprises the steps of; receiving a parametric equation of a closed planar curve; choosing nodes on the closed planar curve with equal intervals; generating a continuous scale space of the nodes on the curve; calculating circle of curvature for every node on the closed curve for every scale in every octave; finding circle of curvature differences between plurality of adjacent scales; comparing each curvature difference value and choosing the nodes with a minimum or maximum curvature difference as feature points; representing the outline with a descriptor including all the feature points. The method further comprises the steps; eliminating the feature points which are closer to each other than a predetermined threshold; and comparing a descriptor with each previously recorded descriptor belonging to various outlines, finding at least one descriptor with a good match.

Claims

1. A method for identifying scale invariant features of image outlines (100) characterized by the steps of;

receiving a parametric equation of a closed planar curve (101),

choosing nodes on the closed planar curve with equal intervals (102),

generating a continuous scale space of the nodes on the curve by successively convolving the planar curve with a Gaussian filter and down-sampling it after each octave (103),

calculating a value representing the amount of curvature for every node on the closed curve for every scale in every octave (104),

finding curvature differences between plurality of adjacent scales and forming curvature difference groups (105),

comparing each curvature difference value with curvature differences belonging to the same and adjacent groups; and

choosing the nodes with a minimum or maximum curvature difference as feature points (106),

representing the outline with a descriptor having a center representing a highest available scale value of key points as radial arrays in which radial distance from the center represents a scale value of difference groups, rotation around this center represents the outline curve parameter and the value at each point represents the feature orientation and the radial array is encoded on a two dimensional color image and two dimensional vectors representing the feature'"'"'s orientations are encoded with the color of the placed point (107).

2. The method for identifying scale invariant features of image outlines (100) according to claim 1, characterized in that an image pre-processing operation, which generates a binary image from an input image and fits a curve on the boundary of an object is implemented before step (101).

3. The method for identifying scale invariant features of image outlines (100) according to claim 2, characterized in that the planar curve is down-sampled into half in step (103).

4. The method for identifying scale invariant features of image outlines (100) according to claim 2, characterized in that a continuous scale space is generated by convolving the planar curve with a Gaussian filter throughout each octave in step (103).

5. The method for identifying scale invariant features of image outlines (100) according to claim 2, characterized in that a function that outputs inverse of the radius of the circle, which is tangent to the curve at a given node is used to calculate curvature in step (104).

6. The method for identifying scale invariant features of image outlines (100) according to claim 2, characterized in that two neighbours on the current difference group and three closest nodes on the adjacent groups are used for comparison in step (106).

7. The method for identifying scale invariant features of image outlines (100) according to claim 1, characterized in that the planar curve is down-sampled into half in step (103).

8. The method for identifying scale invariant features of image outlines (100) according to claim 7, characterized in that a continuous scale space is generated by convolving the planar curve with a Gaussian filter throughout each octave in step (103).

9. The method for identifying scale invariant features of image outlines (100) according to claim 7, characterized in that a function that outputs inverse of the radius of the circle, which is tangent to the curve at a given node is used to calculate curvature in step (104).

10. The method for identifying scale invariant features of image outlines (100) according to claim 1, characterized in that a continuous scale space is generated by convolving the planar curve with a Gaussian filter throughout each octave in step (103).

11. The method for identifying scale invariant features of claim outlines (100) according to claim 10, characterized in that a function that outputs inverse of the radius of the circle, which is tangent to the curve at a given node is used to calculate curvature in step (104).

12. The method for identifying scale invariant features of image outlines (100) according to claim 1, characterized in that a function that outputs inverse of the radius of the circle, which is tangent to the curve at a given node is used to calculate curvature in step (104).

13. The method for identifying scale invariant features of image outlines (100) according to claim 1, characterized in that two neighbours on the current difference group and three closest nodes on the adjacent groups are used for comparison in step (106).

14. The method for identifying scale invariant features of image outlines (100) according to claim 1, characterized by the step of eliminating the feature points which are closer to each other than a predetermined threshold along the outline (108).

15. The method for identifying scale invariant features of image outlines (100) according to claim 14, characterized in that the threshold is selected as predetermined percent of the entire silhouette outline length in step (108).

16. The method for identifying scale invariant features of image outlines (100) according to claim 1, characterized by the step of comparing a descriptor with each previously recorded descriptor belonging to various outlines and finding at least one descriptor with a good match (109).

17. The method for identifying scale invariant features of image outlines (100) according to claim 1, characterized in that descriptor pairs are compared by finding the difference at every possible rotation of radial arrays around the same center in step (109).

18. The method for identifying scale invariant features of image outlines (100) according to claim 17, characterized in that one of the descriptors'"'"' orientation values are cyclically varied for at least one relative rotation of radial arrays in step (109).

19. The method for identifying scale invariant features of image outlines (100) according to claim 1, characterized in that pairs giving the least difference upon comparison will give the best match in step (109).

20. A system for identifying scale invariant features of image outlines (1) characterized by;

at least one feature and image database (2) configured to store outline descriptors determined by steps (101) to (108) as claimed in any of the preceding claims and respective outlines together with original pixel images,

at least one processing unit (3) connected to feature and image database (1) and configured to receive a parametric equation of an outline and implement the method for identifying scale invariant features of image outlines (100) as claimed in claim 1, using this outline as the input.