Local feature descriptor extracting apparatus, method for extracting local feature descriptor, and program

• US 9,613,287 B2
• Filed: 11/15/2012
• Issued: 04/04/2017
• Est. Priority Date: 11/18/2011
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Iwamoto, Kota, Mase, Ryota
• Primary Examiner(s): STREGE, JOHN B
• Application Number:
  US14/359,049
• Publication Number:
  US 20140321755A1
• Time in Patent Office: 1,601 Days
• CPC Class Codes: G06F 18/2135 based on approximation criteria, e.g. principal component analysis
• CPC Class Codes: G06V 10/462 Salient features, e.g. scale invariant feature transforms [SIFT]
• 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
• CPC Class Codes: G06V 10/7715 Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods

Abstract

The size of a feature descriptor is reduced with the accuracy of object identification maintained. A local feature descriptor extracting apparatus includes a feature point detecting unit configured to detect feature points in an image, a local region acquiring unit configured to acquire a local region for each of the feature points, a subregion dividing unit configured to divide each local region into a plurality of subregions, a subregion feature vector generating unit configured to generate a feature vector with a plurality of dimensions for each of the subregions in each local region, and a dimension selecting unit configured to select dimensions from the feature vector in each subregion so as to reduce a correlation between the feature vectors in proximate subregions based on positional relations among the subregions in each local region and output elements of the selected dimensions as a feature descriptor of the local region.

Claims

1. A local feature descriptor extracting apparatus comprising:

memory that stores an image;

a processor configured to;

detect feature points in the image;

acquire a first local region for each of the feature points;

divide each first local region into a first plurality of subregions;

generate a first feature vector with a plurality of dimensions for each of the subregions in each first local region;

select dimensions from the first feature vector in each subregion such that at least one gradient orientation differs between proximate subregions and output elements of the selected dimensions as a feature descriptor of the first local region; and

store, in association with the image, the feature descriptor of each first local region of the image in the memory.

2. The local feature descriptor extracting apparatus according to claim 1, wherein the processor is configured to select the dimensions from the first feature vector in such a manner that the adjacent subregions are different from each other in at least one of the selected dimensions.

3. The local feature descriptor extracting apparatus according to claim 1, wherein the processor is configured to select the dimensions from the first feature vector in accordance with a selection order for selecting the dimensions from the first feature vectors in the first plurality of subregions in the first local region.

4. The local feature descriptor extracting apparatus according to claim 3, wherein the processor is configured to output elements of the selected dimensions in order of the dimensions selected in accordance with the selection order, as the feature descriptor of the first local region.

5. The local feature descriptor extracting apparatus according to claim 1, wherein the processor is configured to determine the number of dimensions selected and to select the dimensions from the first feature vector in such a manner that the number of dimensions for the feature descriptor of the first local region is equal to the determined number of dimensions.

6. The local feature descriptor extracting apparatus according to claim 5, wherein the processor is configured to determine the number of dimensions selected for the feature descriptor of the first local region for each feature point, according to the number of the detected feature points.

7. The local feature descriptor extracting apparatus according to claim 5, wherein the processor is configured to determine the number of dimensions selected for the feature descriptor of the first local region for each feature point, based on information indicative of importance of each feature point.

8. The local feature descriptor extracting apparatus according to claim 1, wherein the processor is configured to quantize the first feature vector in each subregion, and

to select the dimensions from the quantized first feature vector.

9. The local feature descriptor extracting apparatus according to claim 1, wherein the processor is configured to quantize the feature descriptor of the first local region.

10. The local feature descriptor extracting apparatus according to claim 1, wherein the processor is configured to generate, based on the first feature vectors in a plurality of proximate subregions, a feature vector for an expanded subregion that is a subregion larger than each of the plurality of proximate subregions, and

to select dimensions from the feature vector in each expanded subregion to reduce a correlation between the proximate expanded subregions based on positional relations among the expanded subregions in each first local region and output elements of the selected dimensions as a feature descriptor of the first local region.

11. The local feature descriptor extracting apparatus according to claim 10, wherein the elements of the first feature vector are a feature descriptor depending on an orientation, and

the processor is configured to generate a feature vector for the expanded subregion based on elements of dimensions, of the first feature vector in each subregion, which are not selected.

12. The local feature descriptor extracting apparatus according to claim 1, wherein the processor is configured to generate, for each of the subregions in each first local region, a second feature vector with a plurality of dimensions which is different, in orientation of at least one element, from the first feature vector.

13. The local feature descriptor extracting apparatus according to claim 12, wherein the processor is configured to divide each first local region into a second plurality of subregions different from the first plurality of subregions, and to generate the feature vector for each of the subregions of the second plurality of subregions.

14. The local feature descriptor extracting apparatus according to claim 13, wherein the processor is configured to acquire, for each feature point, a second local region different from the first local region acquired by the processor, and to divide each of the second local regions into a plurality of subregions.

15. A method for extracting a local feature descriptor, the method performed by a processor in communication with memory comprising:

retrieving an image from the memory;

detecting feature points in an image;

acquiring a local region for each of the feature points;

dividing each local region into a plurality of subregions;

generating a feature vector with a plurality of dimensions for each of the subregions in each local region;

selecting dimensions from the feature vector in each subregion such that at least one gradient orientation differs between proximate subregions and outputting elements of the selected dimensions as a feature descriptor of the local region; and

storing, in association with the image, the feature descriptor of each local region of the image in the memory.

16. A non-transitory computer-readable storage medium storing a program for causing a computer to implement:

a function of retrieving an image from memory;

a function of detecting feature points in the image;

a function of acquiring a local region for each of the feature points;

a function of dividing each local region into a plurality of subregions;

a function of generating a feature vector with a plurality of dimensions for each of the subregions in each local region;

a function of selecting dimensions from the feature vector in each subregion such that at least one gradient orientation differs between proximate subregions and outputting elements of the selected dimensions as a feature descriptor of the local region; and

a function of storing, in association with the image, the feature descriptor of each local region of the image in the memory.