Method and apparatus for searching an image, and computer-readable recording medium for executing the method

Patent metadata

Field	Value
Publication number	US-9747308-B2
Application number	US-201214343164-A
Country	US
Kind code	B2
Filing date	Nov 12, 2012
Priority date	Nov 16, 2011
Publication date	Aug 29, 2017
Grant date	Aug 29, 2017

Abstract

The present disclosure relates to a method and apparatus for searching an image, and to a computer-readable recording medium for executing the method. The apparatus for searching an image of the present disclosure obtains features of an input image; and obtains words that correspond to the features respectively and an adjacent word that is adjacent to the words corresponding to the features. When a word is assigned to a first word cell of a plurality of word cells that are included in a visual feature space, an adjacent word is assigned to at least one second word cell that is adjacent to the first word cell, where the plurality of word cells is assigned to different words, and at least one word being within a predetermined distance from a word is designated as the adjacent word. The apparatus is further configured to search for an image that is identical or similar to the input image based on information associated with a first group of images corresponding to the word and information associated with a second group of images corresponding to the adjacent word, the information on the first and second groups of images being stored in a database.

First claim

The invention claimed is:

1. An image searching method, comprising:
(a) obtaining, by an image search apparatus, feature descriptors of an input image;
(b) obtaining, by the image search apparatus, words to which the respective features correspond, and adjacent words that are adjacent to the respective words to which the features correspond, wherein when a first word cell of a plurality of word cells that are included in a visual feature space is assigned to a word, an adjacent word is a word assigned to at least one second word cell of the plurality of word cells, the second word cell being adjacent to the first word cell to which the word is assigned, wherein the plurality of word cells is assigned to different words, and at least one word being within a predetermined distance from the word is designated as the adjacent word, wherein the plurality of word cells in the visual feature space is formed by obtaining features of a plurality of respective images stored in the database, and clustering the features of the plurality of images based on the locations in the visual feature space, and the plurality of word cells is assigned to different words, and further wherein at least one of the words is obtained by using information on center points, each of which is a mean value of features of images that correspond to the respective word cell, to compare distances between the respective features of the input image and the respective center points of the plurality of word cells, and obtaining, for the word, a word assigned to a word cell of which center point is most closely adjacent to the corresponding feature of the input image;
(c) searching for, by the image search apparatus, an image that is identical or similar to the input image based on information associated with a first group of images corresponding to the word and information associated with a second group of images corresponding to the adjacent word, the information on the first and second groups of images being stored in a database, wherein the information on the first group of images corresponding to the word includes identification symbols and feature descriptors of the respective first group of images corresponding to the first word cell to which the word is assigned from among the plurality of word cells in the visual feature space, and the information on the second group of images corresponding to the adjacent word includes identification symbols and feature descriptors of the respective second group of images corresponding to at least one second word cell adjacent to the first word cell to which the word is assigned from among the plurality of word cells in the visual feature space.

2. The method of claim 1 , wherein the distance information is information on a Euclidian distance.

3. The method of claim 1 , wherein step (c) comprises:
allocating an identification symbol of at least one image of the first and second groups of images to each of the features of the input image based on distances of the respective feature of the input image from the respective features of the first and second groups of images; and
identifying an image that is identical or similar to the input image based on the identification symbols allocated to the respective features of the input image.

4. An image search apparatus, comprising:
a processor to:
obtain features of an input image;
obtain words to which the respective features correspond, and adjacent words that are adjacent to the words, wherein when a first word cell of a plurality of word cells that are included in a visual feature space is assigned to a word, an adjacent word is a word assigned to at least one second word cell of the plurality of word cells, the second word cell being adjacent to the first word cell to which the word is assigned, and the plurality of word cells being assigned to different words, and at least one word being within a predetermined distance from the word is designated as the adjacent word, wherein the plurality of word cells in the visual feature space is formed by obtaining features of a plurality of respective images stored in the database, and clustering the features of the plurality of images based on the locations in the visual feature space, and the plurality of word cells is assigned to different words, and further wherein at least one of the words is obtained by using information on center points, each of which is a mean value of features of images that correspond to the respective word cell, to compare distances between the respective features of the input image and the respective center points of the plurality of word cells, and obtaining, for the word, a word assigned to a word cell of which center point is most closely adjacent to the corresponding feature of the input image; and
search for an image that is identical or similar to the input image based on information associated with a first group of images corresponding to the word and information associated with a second group of images corresponding to the adjacent word, the information on the first and second groups of images being stored in a database, wherein the information on the first group of images corresponding to the word includes identification symbols and feature descriptors of the respective first group of images corresponding to the first word cell to which the word is assigned from among the plurality of word cells in the visual feature space, and the information on the second group of images corresponding to the adjacent word includes identification symbols and feature descriptors of the respective second group of images corresponding to at least one second word cell adjacent to the first word cell to which the word is assigned from among the plurality of word cells in the visual feature space.

5. The apparatus of claim 4 , further comprising a lookup table including information on the adjacent words mapped to the words in the visual feature space.

6. The apparatus of claim 4 , wherein the processor further to:
allocate an identification symbol of at least one image of the first group of images and the second group of images to each of the features of the input image based on distances of the respective feature of the input image to the respective features of the first and second groups of images; and
identify an image that is identical or similar to the input image based on the identification symbols allocated to the respective features of the input image.

7. A non-transitory computer-readable recording medium for recording a computer program comprising instructions, which when executed by a processor, causes the processor to perform operations comprising:
(a) obtaining, by an image search apparatus, features of an input image;
(b) obtaining, by the image search apparatus, words to which the respective features correspond, and adjacent words that are adjacent to the respective words to which the features correspond, wherein when a first word cell of a plurality of word cells that are included in a visual feature space is assigned to a word, an adjacent word is a word assigned to at least one second word cell of the plurality of word cells, the second word cell being adjacent to the first word cell to which the word is assigned, wherein the plurality of word cells is assigned to different words, and at least one word being within a predetermined distance from the word is designated as the adjacent word, wherein the plurality of word cells in the visual feature space is formed by obtaining features of a plurality of respective images stored in the database, and clustering the features of the plurality of images based on the locations in the visual feature space, and the plurality of word cells is assigned to different words, and further wherein at least one of the words is obtain

Assignees

• Kim Tae-Hoon
• Park Minje
• Choi Songki
• Intel Corp

Inventors

• Kim Tae-Hoon
• Park Minje
• Choi Songki

Classifications

• G06V10/768

  using context analysis, e.g. recognition aided by known co-occurring patterns · CPC title

• G06V10/467

  Encoded features or binary features, e.g. local binary patterns [LBP] · CPC title

• G06F17/30247

  Physics · mapped topic

• G06F17/30268 Primary

  Physics · mapped topic

• G06K9/72

  Physics · mapped topic