Image Clustering Method, Image Clustering System, And Image Clustering Server

Patent metadata

Field	Value
Publication number	US-2018239986-A1
Application number	US-201815957276-A
Country	US
Kind code	A1
Filing date	Apr 19, 2018
Priority date	Nov 14, 2013
Publication date	Aug 23, 2018
Grant date	—

Abstract

A method and a clustering system for image clustering, and a computer-readable storage medium are provided. The method includes: extracting a GIST feature of a first image and a GIST feature of a second image; obtaining an image fingerprint of the first image, based on the GIST feature of the first image and in conjunction with an LSH algorithm and obtaining an image fingerprint of the second image, based on the GIST feature of the second image and in conjunction with the LSH algorithm; calculating a similarity between the first and second images, based on the image fingerprints of the first and second images; and classifying the first image and the second image as a same category of image in a case that the similarity between the first image and the second image is larger than a predetermined similarity threshold.

First claim

1 . An image clustering method, comprising:
extracting a Global Scene Semantic (GIST) feature of a first image and a GIST feature of a second image; obtaining an image fingerprint of the first image, based on the GIST feature of the first image and in conjunction with a Local Sensitive Hashing (LSH) algorithm and obtaining an image fingerprint of the second image, based on the GIST feature of the second image and in conjunction with the LSH algorithm; calculating a similarity between the first image and the second image, based on the image fingerprint of the first image and the image fingerprint of the second image; and classifying the first image and the second image as a same category of image in a case that the similarity between the first image and the second image is larger than a predetermined similarity threshold.

2 . The method according to claim 1 , wherein the first image is a first advertising image comprised in a first advertising order and the second image is a second advertising image comprised in a second advertising order.

3 . The method according to claim 2 , further comprising:
collecting the first advertising order and the second advertising order, wherein the first advertising order comprises a first advertising order identification and the first advertising image carrying a first Uniform Resource Locater (URL), and the second advertising order comprises a second advertising order identification and the second advertising image carrying a second URL.

4 . The method according to claim 1 , wherein
obtaining an image fingerprint of an image, based on the GIST feature of the image and in conjunction with a Local Sensitive Hashing (LSH) algorithm comprises: initializing an f-dimensional GIST feature vector and an f-bit binary as 0; for each dimension of GIST feature of the image: generating an f-bit signature for the GIST feature with a conventional hash algorithm; in a case that an i-th bit of the f-bit signature is 1, adding an i-th element of the f-dimensional GIST feature vector with a weight of the dimension of GIST feature; and in a case that the i-th bit of the f-bit signature is not 1, subtracting the weight of the dimension of GIST feature from the i-th element of the f-dimensional GIST feature vector, where 1≤i≤f; adding the f-dimensional GIST feature vectors for all dimensions of GIST feature of the image to obtain a first f-dimensional GIST feature vector; and outputting the f-bit binary as an image fingerprint of the image, wherein an i-th bit of the f-bit binary is 1 if an i-th element of the first f-dimensional GIST feature vector is larger than 0; and an i-th bit of the f-bit binary is 0 if an i-th element of the first f-dimensional GIST feature vector is not larger than 0.

5 . The method according to claim 2 , further comprising: storing the first advertising order and the second advertising order into an advertising order set corresponding to a same cluster identification in a database, in a case that the similarity between the first advertising image and the second advertising image is larger than the predetermined similarity threshold.

6 . The method according to claim 5 , further comprising:
collecting a third advertising order which comprises a third advertising order identification and a third advertising image carrying a third URL; determining whether an advertising image carrying the third URL is stored in the database, wherein the advertising image is included in an advertising order; and in the case that the advertising image carrying the third URL is stored in the database, storing the third advertising order into an advertising order set to which the advertising order belongs.

7 . An image clustering system, comprising a processor and memory configured to store program instructions, when executed by the processor, which cause the processor to perform operations comprising:
extracting a Global Scene Semantic (GIST) feature of a first image and a GIST feature of a second image; obtaining an image fingerprint of the first image, based on the GIST feature of the first image and in conjunction with a Local Sensitive Hashing (LSH) algorithm and obtaining an image fingerprint of the second image, based on the GIST feature of the second image and in conjunction with the LSH algorithm; calculating a similarity between the first image and the second image, based on the image fingerprint of the first image and the image fingerprint of the second image; and classifying the first image and the second image as a same category of image, in the case that the similarity between the first image and the second image is larger than a determined similarity threshold.

8 . The system according to claim 7 , wherein the first image is a first advertising image comprised in a first advertising order, the second image is a second advertising image comprised in a second advertising order, and the operations further comprises:
collecting the first advertising order and the second advertising order, wherein the first advertising order comprises a first advertising order identification and the first advertising image carrying a first Uniform Resource Locater (URL), and the second advertising order comprises a second advertising order identification and the second advertising image carrying a second URL.

9 . The system according to claim 8 , wherein the operations further comprises: storing the first advertising order and the second advertising order into an advertising order set corresponding to the same cluster identification in a database, in the case that the similarity between the first advertising image and the second advertising image is larger than the predetermined similarity threshold.

10 . The system according to claim 9 , wherein is the operations further comprises:
collecting a third advertising order which comprises a third advertising order identification and the third advertising image carrying a third URL;
determining whether an advertising image carrying the third URL is stored in the database, wherein the advertising image is comprised in an advertising order; and
in the case that the adverting image carrying the third URL is stored in the database, storing the third adverting order into an advertising order set to which the adverting order belongs.

11 . The system according to claim 7 , wherein the fingerprint obtaining unit is further programmed to:
initialize an f-dimensional GIST feature vector and an f-bit binary as 0; for each dimension of GIST feature of the image; generate an f-bit signature for the GIST feature with a conventional hash algorithm; in a case that an i-th bit of the f-bit signature is 1, add an i-th element of the f-dimensional GIST feature vector with a weight of the dimension of GIST feature; and in a case that the i-th bit of the f-bit signature is not 1, subtract the weight of the dimension of GIST feature from the i-th element of the f-dimensional GIST feature vector, where 1≤i≤f, add the f-dimensional GIST feature vectors for all dimensions of GIST feature of the image to obtain a first f-dimensional GIST feature vector; and output the f-bit binary as an image fingerprint of the image, wherein an i-th bit of the f-bit binary is 1 if an i-th element of the first f-dimensional GIST feature vector is larger than 0; and an i-th bit of the f-bit binary is 0 if an i-th element of the first f-dimensional GIST feature vector is not larger than 0.

12 . A non-transient computer-readable storage medium storing computer executable instructions which, when run by a computer, cause the following steps to be executed:
extracting a Global Scene Semantic (GIST) feature of a first image and a G

Assignees

• Tencent Tech Shenzhen Co Ltd

Inventors

• Huang Pipei

Classifications

• G06V10/762

  using clustering, e.g. of similar faces in social networks · CPC title

• G06V10/761

  Proximity, similarity or dissimilarity measures · CPC title

• G06F18/23 Primary

  Clustering techniques · CPC title

• G06Q30/0277 Primary

  Online advertisement · CPC title

• G06F18/241

  relating to the classification model, e.g. parametric or non-parametric approaches · CPC title