Method and device for finding nearest neighbor

• US 9,177,227 B2
• Filed: 12/15/2011
• Issued: 11/03/2015
• Est. Priority Date: 12/17/2010
• Status: Active Grant

• Current Assignee: Ivisys Sweden AB
• Original Assignee: Ivisys APS
• Inventors: Chehaiber, Moatasem
• Primary Examiner(s): Park, Chan
• Assistant Examiner(s): TRAN, MAI H
• Application Number:
  US13/990,246
• Publication Number:
  US 20130251269A1
• Time in Patent Office: 1,419 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 16/583 using metadata automatically derived from the content
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06F 18/24147 Distances to closest patterns, e.g. nearest neighbour classification
• CPC Class Codes: G06V 10/462 Salient features, e.g. scale invariant feature transforms [SIFT]
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects

Abstract

The present invention relates to a method and a device for finding nearest neighbor. In particular, it relates to a sorting, searching and matching multiple dimensional data, such as vectors, in order to find the nearest neighbor. The method is particularly useful as part of a SIFT algorithm.

Claims

1. A method comprising:

identifying scale invariant features in at least one image by matching a specific vector among a dataset consisting of a plurality of vectors, wherein the vectors represent scale invariant features, comprising;

i. Selecting a reference point vector, wherein the reference point vector is selected among the data set;

ii. Calculating the distances, d, between said reference point vector and each of the vectors of the data set;

iii. Re-arranging the plurality of vectors of the data set as a function of ascending distance, d, to the reference point vector;

iv. Sorting the vectors of the data set into groups of vectors having the same distance, d_group, from said reference point vector;

v. Subsequently rearranging the vectors of each of the groups of vectors, which comprises more than two vectors, such that the second vector of the group has a minimum distance to the first vector of the group, and each subsequent vector of the group has a minimum distance to the previous vector of the group;

vi. Identifying the best match for said specific vector by;

a. Calculating the distance, d_specific, between said reference point vector and the specific vector;

b. Comparing the calculated distance, d_specific, and a distance between the reference point vector and a vector positioned in the middle of the re-arranged plurality of vectors;

c. Identifying, using successive approximation and based on the distance between the reference point vector and a vector positioned in the middle of the re-arranged plurality of vectors, the group of vectors having vectors with a distance, d, closest to said calculated distance, d_specific;

d. Identifying the vector or vectors within the identified group or groups having the closest distance, d_minimum, to said specific vector;

e. Identifying any additional group with distances, d, from said reference point vector in an interval from the larger of zero and d_specific−

d_minimum to d_specific+d_minimum; and

f. Repeating steps vi.d. and vi.e. until all groups in said interval have been examined.

2. The method according to claim 1, wherein the vectors are multiple dimensional.

3. The method according to claim 1, wherein said distances, d, d_group, d_specific and d_minimum, are calculated as a distance selected among the group consisting of Euclidean, Mahalanobis, Manhattan, Chebyshe and Minkowski distance.

4. The method according to claim 1, wherein said distances, d, d_group, d_specific and d_minimum, are calculated as Euclidean distance.

9. A non-transitory computer-readable medium storing a software program implementing the method according to claim 1.

10. The non-transitory computer-readable medium according to claim 9, adapted to identify scale invariant features in an image defined by a plurality of pixels.

5. An image processing device, said image processing device comprising a processor circuit configured to:

Identify scale invariant features in an image defined by a plurality of pixels by;

i. Selecting a reference point vector, wherein the reference point vector is selected among the data set;

ii. Calculating the distances, d, between said reference point vector and each of the vectors of the data set;

iii. Re-arranging the plurality of vectors of the data set as a function of ascending distance, d, to the reference point vector;

iv. Sorting the vectors of the data set into groups of vectors having the same distance, d_group, from said reference point vector;

v. Subsequently rearranging the vectors of each of the groups of vectors, which comprises more than two vectors, such that the second vector of the group has a minimum distance to the first vector of the group, and each subsequent vector of the group has a minimum distance to the previous vector of the group;

vi. Identifying the best match for said specific vector by;

a. Calculating the distance, d_specific, between said reference point vector and the specific vector;

b. Comparing the calculated distance, d_specific, and a distance between the reference point vector and a vector positioned in the middle of the re-arranged plurality of vectors;

c. Identifying, using successive approximation and based on the distance between the reference point vector and a vector positioned in the middle of the re-arranged plurality of vectors, the group of vectors having vectors with a distance, d, closest to said calculated distance, d_specific;

d. Identifying the vector or vectors within the identified group or groups having the closest distance, d_minimum, to said specific vector;

e. Identifying any additional group with distances, d, from said reference point vector in an interval from the larger of zero and d_specific−

d_minimum to d_specific+d_minimum, and

f. Repeating steps vi.d. and vi.e. until all groups in said interval have been examined.

6. The image processing device according to claim 5, said image processing device comprising multiple chips, each chip configured to perform part of the identification of the best match for the specific vector of step vi.

7. The image processing device according to claim 5, said image processing device comprising at least one element selected among the group consisting of a GPU, DSP, FPGA and an ASIC.

8. The image processing device according to claim 5, said image processing device comprising:

means for storing the reference point vector, the sorted vectors of the data set, the distances, d and d_group; and

means for receiving the specific vector.