Image search device, image search method, program, and computer-readable storage medium

1. An image search device, comprising:

a plurality of parallel processors each of which includes a register and which execute a same instruction;

a common memory which is commonly and directly accessible from the plurality of parallel processors,

wherein each element of a vector stored in common memory are loaded, at one time, into the register of a corresponding one of the plurality of parallel processors in bulk, and each of the plurality of parallel processors processes the loaded element of the vector;

a storage unit storing a plurality of image feature vectors that respectively indicate a plurality of local features of each of a plurality of search target images to be searched and respectively belong to one of a plurality of clusters, and a plurality of representative feature vectors each of which represents one of the plurality of clusters, in which more than one of the image feature vectors are extracted from one of the plurality of images;

representative vector transfer means for transferring the plurality of the representative feature vectors from the storage unit to the common memory;

query feature obtaining means for obtaining and storing, in the common memory, a plurality of query feature vectors which are created from one image serving as a query, each query feature vector indicating a local feature of the image serving as the query;

first distance calculating means for calculating a first distance between at least a part of the plurality of transferred representative feature vectors and each of the query feature vectors using the plurality of parallel processors;

cluster selecting means for selecting a cluster from the plurality of clusters based on the calculated first distance;

second distance calculating means for calculating a second distance between the query feature vector and the plurality of image feature vectors which belong to the selected cluster using the plurality of parallel processors;

image feature vector selecting means for selecting, for each of the plurality of query feature vectors, an image feature vector from the plurality of image feature vectors based on the calculated second distance, where, for any one of plurality of the query feature vectors, a selected image feature vector is not selected relative to other query feature vectors within the plurality of query feature vectors;

score generating means for generating an image score of the search target image based on a total sum of score elements each score element depending on proximities between the selected plurality of image feature vectors each indicating a feature of any one of the plurality of search target images and the query feature vectors corresponding to the selected image feature vectors; and

image selecting means for selecting at least one of the plurality of search target images to be searched based on the image score.

2. The image search device according to claim 1 , wherein the image feature vector selecting means selects, for each of the plurality of query feature vectors, an image feature vector whose distance from the plurality of query feature vectors is smaller than a predetermined threshold value from among a plurality of image feature vectors which respectively indicate a plurality of local features of each of a plurality of search target images to be searched and which are stored in the storage unit in advance.

3. The image search device according to claim 2 , wherein the image feature vector selecting means selects, for each of the plurality of query feature vectors, not more than one image feature vector is from among a plurality of image feature vectors which respectively indicate a plurality of local features of each of a plurality of search target images to be searched and which are stored in the storage unit in advance.

4. The image search device according to claim 3 ,

wherein a cluster corresponding to each of the plurality of query feature vectors is selected from a plurality of clusters into which the plurality of image feature vectors stored in the storage unit are classified,

wherein, in the image feature vector selection, the image feature vector is selected for each of the plurality of query feature vectors from the image feature vectors belonging to the selected cluster.

5. The image search device according to claim 2 , further comprising:

wherein a cluster corresponding to each of the plurality of query feature vectors is selected from a plurality of clusters into which the plurality of image feature vectors stored in the storage unit are classified,

wherein, in the image feature vector selection, the image feature vector is selected for each of the plurality of query feature vectors from the image feature vectors belonging to the selected cluster.

6. The image search device according to claim 2 , wherein the proximity between the selected image feature vectors and the query feature vector corresponding to the selected image feature vector is a distance between the selected image feature vectors and the query feature vector.

7. The image search device according to claim 1 , wherein the image feature vector selecting means selects, for each of the plurality of query feature vectors, an image feature vector of which a number is predetermined in ascending order of distance from among a plurality of image feature vectors which respectively indicate a plurality of local features of each of a plurality of search target images to be searched and which are stored in the storage unit in advance.

8. The image search device according to claim 7 , wherein the image feature vector selecting means selects, for each of the plurality of query feature vectors, not more than one image feature vector from among a plurality of image feature vectors which respectively indicate a plurality of local features of each of a plurality of search target images to be searched and which are stored in the storage unit in advance.

9. The image search device according to claim 8 ,

wherein a cluster corresponding to each of the plurality of query feature vectors is selected from a plurality of clusters into which the plurality of image feature vectors stored in the storage unit are classified,

wherein, in the image feature vector selection, the image feature vector is selected for each of the plurality of query feature vectors from the image feature vectors belonging to the selected cluster.

10. The image search device according to claim 7 ,

wherein a cluster corresponding to each of the plurality of query feature vectors is selected from a plurality of clusters into which the plurality of image feature vectors stored in the storage unit are classified,

wherein, in the image feature vector selection, the image feature vector is selected for each of the plurality of query feature vectors from the image feature vectors belonging to the selected cluster.

11. The image search device according to claim 7 , wherein the proximity between the selected image feature vectors and the query feature vector corresponding to the selected image feature vector is a distance between the selected image feature vectors and the query feature vector.

12. The image search device according to claim 1 , wherein the image feature vector selecting means selects, for each of the plurality of query feature vectors, not more than one image feature vector from among a plurality of image feature vectors which respectively indicate a plurality of local features of each of a plurality of search target images to be searched and which are stored in the storage unit in advance.

13. The image search device according to claim 12 ,

wherein a cluster corresponding to each of the plurality of query feature vectors is selected from a plurality of clusters into which the plurality of image feature vectors stored in the storage unit are classified,

wherein, in the image feature vector selection, the image feature vector is selected for each of the plurality of query feature vectors from the image feature vectors belonging to the selected cluster.

14. The image search device according to claim 12 , wherein the proximity between the selected image feature vectors and the query feature vector corresponding to the selected image feature vector is a distance between the selected image feature vectors and the query feature vector.

15. The image search device according to claim 1 ,

wherein a cluster corresponding to each of the plurality of query feature vectors is selected from a plurality of clusters into which the plurality of image feature vectors stored in the storage unit are classified,

wherein, in the image feature vector selection, the image feature vector is selected for each of the plurality of query feature vectors from the image feature vectors belonging to the selected cluster.

16. The image search device according to claim 15 , wherein the proximity between the selected image feature vectors and the query feature vector corresponding to the selected image feature vector is a distance between the selected image feature vectors and the query feature vector.

17. The image search device according to claim 1 , wherein the proximity between the selected image feature vectors and the query feature vector corresponding to the selected image feature vector is a distance between the selected image feature vectors and the query feature vector.

18. An image search method, which uses

a plurality of parallel processors each of which includes a register and which execute a same instruction;

a common memory which is commonly and directly accessible from the plurality of parallel processors,

wherein each element of a vector stored in common memory are loaded, at one time, into the register of a corresponding one of the plurality of parallel processors in bulk, and each of the plurality of parallel processors processes the loaded element of the vector;

a storage unit storing a plurality of image feature vectors that respectively indicate a plurality of local features of each of a plurality of search target images to be searched and respectively belong to one of a plurality of clusters, and a plurality of representative feature vectors each of which represents one of the plurality of clusters, in which more than one of the image feature vectors are extracted from one of the plurality of images;

representative vector transfer means for transferring the plurality of the representative feature vectors from the storage unit to the common memory;

said method comprising:

obtaining and storing, in a common memory, a plurality of query feature vectors which are created from one image serving as a query, each query feature vector indicating a local feature of the image serving as the query;

calculating a first distance between at least a part of the plurality of transferred representative feature vectors and each of the query feature vectors using the plurality of parallel processors;

selecting a cluster from the plurality of clusters based on the calculated first distance;

calculating a second distance between the query feature vector and the plurality of image feature vectors which belong to the selected cluster using the plurality of parallel processors;

selecting, for each of the plurality of query feature vectors, an image feature vector from the plurality of image feature vectors based on the calculated second distance, where, for any one of plurality of the query feature vectors, a selected image feature vector is not selected relative to other query feature vectors within the plurality of query feature vectors;

generating an image score of the search target image based on a total sum of score elements each depending on proximities between the selected plurality of image feature vectors each indicating a feature of any one of the plurality of search target images and the query feature vectors corresponding to the selected image feature vectors; and

selecting at least one of the plurality of search target images to be searched based on the image score.

19. A computer-readable non-transitory storage medium having stored thereon a program for causing hardware comprising:

a plurality of parallel processors, each of which includes a register and which execute a same instruction:

a common memory, which is commonly and directly accessible from the plurality of parallel processors, wherein each element of a vector stored in common memory are loaded, at one time, into the register of a corresponding one of the plurality of parallel processors in bulk, and each of the plurality of parallel processors processes the loaded element of the vector;

a storage unit, storing a plurality of image feature vectors that respectively indicate a plurality of local features of each of a plurality of search target images to be searched and respectively belong to one of a plurality of clusters, and a plurality of representative feature vectors each of which represents one of the plurality of clusters, in which one more than one of the image feature vectors are extracted from one of the plurality of images;

said computer readable non-transitory storage medium causing said hardware to:

transfer the plurality of the representative feature vectors from the storage unit to the common memory;

obtain and store, in common memory, a plurality of query feature vectors which are created from one image serving as a query, each query feature vector indicating a local feature of an image serving as a query;

calculate a first distance between at least a part of the plurality of transferred representative feature vectors and each of the query feature vectors using the plurality of parallel processors;

select a cluster from the plurality of clusters based on the calculated first distance;

calculate a second distance between the query feature vector and the plurality of image feature vectors which belong to the selected cluster using the plurality of parallel processors;

select, for each of the plurality of query feature vectors, an image feature vector from the plurality of image feature vectors based on the calculated second distance, where, for any one of plurality of the query feature vectors, a selected image feature vector is not selected relative to other query feature vectors within the plurality of query feature vectors;

generate an image score of the search target image based on a total sum of score elements each score element depending on proximities between the selected plurality of image feature vectors and the query feature vectors corresponding to the selected image feature vectors; and

select at least one of the plurality of search target images to be searched based on the image score.