Flexible computer vision

• US 8,965,130 B2
• Filed: 11/09/2011
• Issued: 02/24/2015
• Est. Priority Date: 11/09/2010
• Status: Active Grant

• Current Assignee: BAR ILAN University
• Original Assignee: BAR ILAN University
• Inventors: Kaminka, Gal, Sadeh-Or, Eran
• Primary Examiner(s): Chen, Wenpeng
• Application Number:
  US13/292,554
• Publication Number:
  US 20120121130A1
• Time in Patent Office: 1,203 Days
• Field of Search: None
• US Class Current: 382/195
• CPC Class Codes: G06V 10/462 Salient features, e.g. scale invariant feature transforms [SIFT]

Abstract

Methods and apparatus for image matching using local features, in particular a method and apparatus for flexible interest point computation. The method involves producing multiple octaves of a digital image, wherein each octave of said multiple scale octaves comprises multiple layers; initiating a process comprising detection and description of interest points, wherein said process is programmed to progress layer-by-layer over said multiple layers of each of said multiple octaves, and to continue to a next octave of said multiple octaves upon completion of all layers of a current octave of said multiple octaves; upon the detection and the description of each interest point of said interest points during said process, recording an indication associated with said interest point in a memory, such that said memory accumulates indications during said process; and upon interruption to said process, returning a result being based at least on said indications.

Claims

1. A method for flexible interest point computation, comprising:

producing multiple octaves of a digital image, wherein each octave of said multiple scale octaves comprises multiple layers;

initiating a process comprising detection and description of interest points, wherein said process is programmed to progress layer-by-layer over said multiple layers of each of said multiple octaves, and to continue to a next octave of said multiple octaves upon completion of all layers of a current octave of said multiple octaves;

upon the detection and the description of each interest point of said interest points during said process, recording an indication associated with said interest point in a memory, such that said memory accumulates indications during said process; and

upon interruption to said process, returning a result being based at least on said indications.

2. The method according to claim 1, wherein:

said indications comprise interest point descriptors computed in said description; and

said returning of the result comprises matching said interest point descriptors with predetermined, comparative interest point descriptors, and returning a result of said matching.

3. The method according to claim 1, wherein:

said process further comprises matching interest point descriptors computed in said description with predetermined, comparative interest point descriptors; and

said indication comprises an indication of a match determined in the matching.

4. The method according to claim 1, further comprising producing multiple octaves of a comparative digital image, wherein each octave of said multiple octaves of said comparative digital image comprises multiple layers,

wherein said detection and description of interest points by said process comprises detection and description of interest points also in said multiple octaves of said comparative digital image, and

wherein said process further comprises matching interest point descriptors of said digital image and of said comparative digital image computed in said description, and

wherein said indication comprises an indication of a match determined in the matching.

5. The method according to claim 1, further comprising producing multiple octaves of a comparative digital image, wherein each octave of said multiple octaves of said comparative digital image comprises multiple layers,

wherein said detection and description of interest points by said process comprises detection and description of interest points also in said multiple octaves of said comparative digital image,

wherein said indications comprise interest point descriptors of said digital image and of said comparative digital image computed in said description, and

wherein said returning of the result comprises matching said interest point descriptors of said digital image with said interest point descriptors of said comparative digital image, and returning a result of said matching.

6. The method according to claim 1, further comprising computing an integral image of said digital image.

7. The method according to claim 1, wherein said process is further programmed to progress from a finest octave of said multiple octaves towards a most coarse octave of said multiple octaves.

8. The method according to claim 1, wherein said process is further programmed to progress from a most coarse octave of said multiple octaves towards a finest octave of said multiple octaves.

9. The method according to claim 8, further comprising computing a homography between said digital image and a comparative digital image.

10. The method according to claim 1, further comprising a stereoscopy of said digital image and of a comparative digital image.

11. The method according to claim 1, wherein said process further comprises computing a Hessian matrix for each octave of said multiple octaves.

12. An apparatus comprising:

a digital imaging device;

a processing unit configured to;

produce multiple octaves of a digital image, wherein each octave of said multiple scale octaves comprises multiple layers;

initiate a process comprising detection and description of interest points, wherein said process is programmed to progress layer-by-layer over said multiple layers of each of said multiple octaves, and to continue to a next octave of said multiple octaves upon completion of all layers of a current octave of said multiple octaves;

upon the detection and the description of each interest point of said interest points during said process, record an indication associated with said interest point in a memory, such that said memory accumulates indications during said process; and

upon interruption to said process, return a result being based at least on said indications.

13. The apparatus according to claim 12, wherein:

said indications comprise interest point descriptors computed in said description; and

said return of the result comprises matching said interest point descriptors with predetermined, comparative interest point descriptors, and returning a result of said matching.

14. The apparatus according to claim 12, wherein:

said process further comprises matching interest point descriptors computed in said description with predetermined, comparative interest point descriptors; and

said indication comprises an indication of a match determined in the matching.

15. The apparatus according to claim 12, wherein said processing unit is further configured to produce multiple octaves of a comparative digital image, wherein each octave of said multiple octaves of said comparative digital image comprises multiple layers,

wherein said detection and description of interest points by said process comprises detection and description of interest points also in said multiple octaves of said comparative digital image, and

wherein said process further comprises matching interest point descriptors of said digital image and of said comparative digital image computed in said description, and

wherein said indication comprises an indication of a match determined in the matching.

16. The apparatus according to claim 12, wherein said processing unit is further configured to produce multiple octaves of a comparative digital image, wherein each octave of said multiple octaves of said comparative digital image comprises multiple layers,

wherein said detection and description of interest points by said process comprises detection and description of interest points also in said multiple octaves of said comparative digital image,

wherein said indications comprise interest point descriptors of said digital image and of said comparative digital image computed in said description, and

wherein said returning of the result comprises matching said interest point descriptors of said digital image with said interest point descriptors of said comparative digital image, and returning a result of said matching.

17. The apparatus according to claim 12, wherein said processing unit is further configured to compute an integral image of said digital image.

18. The apparatus according to claim 12, wherein said process is further programmed to progress from a finest octave of said multiple octaves towards a most coarse octave of said multiple octaves.

19. The apparatus according to claim 12, wherein said process is further programmed to progress from a most coarse octave of said multiple octaves towards a finest octave of said multiple octaves.

20. The apparatus according to claim 12 further comprising a mechanical actuator, wherein said processing unit is further configured, upon interruption to said process, to transmit a control signal based on said result to said mechanical actuator.

21. The apparatus according to claim 20, wherein said mechanical actuator is selected from the group consisting of:

a robotic arm, a motor configured to drive a vehicle and a motor configured to drive a conveyor belt.

22. A method for flexible object recognition, comprising:

receiving predetermined data being visually characteristic of one or more objects;

receiving a digital image;

initiating an analysis of said digital image through consecutive application of filters of different magnitudes to said digital image, to produce, during each filter application of said consecutive applications, at least one indication of correspondence between said digital image and said one or more visual objects; and

interrupting said analysis on demand, and providing a result comprising said at least one indication of correspondence.