Image classification method and device

Patent metadata

Field	Value
Publication number	US-10282597-B2
Application number	US-201615359192-A
Country	US
Kind code	B2
Filing date	Nov 22, 2016
Priority date	Nov 27, 2015
Publication date	May 7, 2019
Grant date	May 7, 2019

Abstract

The present disclosure relates to an image classification and clustering method and device. Two images containing human faces may be classified into one cluster or different clusters based on a cluster merging threshold adaptively determined according to the combinational face poses in the two images. The adaptive cluster merging threshold help reduce false positive and false negative classification of human face containing images into clusters.

First claim

What is claimed is:

1. An image classification method, comprising:
acquiring two images including a first image depicting a first face and a second image depicting a second face;
calculating a similarity between the first face and the second face;
determining the calculated similarity is greater than a preset similarity threshold value;
determining a first face pose information for the first image, the first face pose information including a first orientation value of the first face;
determining a second face pose information for the second image, the second face pose information including a second orientation value of the second face, wherein both the first face pose information and the second face pose information are selected from a pre-established face pose classification mode;
determining, based on the first face pose information and the second face pose information, an angular face relationship between the first face and the second face;
determining an adaptive target cluster merging threshold value based on the angular face relationship; and
clustering the first image and the second image into a same, or different, cluster according to the determined adaptive target cluster merging threshold value.

2. The method of claim 1 , wherein determining the adaptive target cluster merging threshold value comprises:
acquiring a preset cluster merging threshold value;
determining the angular face relationship indicates that the first image and the second image both contain side-view face poses; and
increasing the preset cluster merging threshold value and using the increased preset cluster merging threshold value as the adaptive target cluster merging threshold value.

3. The method of claim 1 , wherein determining the adaptive target cluster merging threshold value comprises:
acquiring a preset cluster merging threshold value;
determining the angular face relationship indicates that the first image and the second image contain face poses of different angles; and
decreasing the preset cluster merging threshold value and using the decreased preset cluster merging threshold value as the adaptive target cluster merging threshold value.

4. The method of claim 1 , wherein determining the adaptive target cluster merging threshold value comprises:
acquiring a preset cluster merging threshold value;
determining the angular face relationship indicates that the first image and the second image both contain front view face poses; and
using the preset cluster merging threshold value as the adaptive target cluster merging threshold value.

5. The method of claim 1 , wherein clustering the first image and the second image comprises:
calculating a cluster merging value;
determining the calculated cluster merging value is greater than the adaptive target cluster merging threshold value; and
determining that the first image and the second image belong to a same cluster.

6. A terminal, comprising:
a processor; and
a memory in communication to the processor for storing instructions executable by the processor;
wherein the processor is configured to:
acquire two images including a first image depicting a first face and a second image depicting a second face;
calculate a similarity between the first face and the second face;
determine the calculated similarity is greater than a preset similarity threshold value;
determine a first face pose information for the first image, the first face pose information including a first orientation value of the first face;
determine a second face pose information for the second image, the second face pose information including a second orientation value of the second face, wherein both the first face pose information and the second face pose information are selected from a pre-established face pose classification mode;
determine, based on the first face pose information and the second face pose information, an angular face relationship between the first face and the second face;
determine an adaptive target cluster merging threshold value based on the angular face relationship; and
cluster the first image and the second image into a same, or different, cluster according to the determined adaptive target cluster merging threshold value.

7. The terminal of claim 6 , wherein, to determine the adaptive target cluster merging threshold value, the processor is further configured to:
acquire a preset cluster merging threshold value;
determine the angular face relationship indicates that the first image and the second image both contain side view face poses; and
increase the preset cluster merging threshold value according and use the increased preset cluster merging threshold value as the adaptive target cluster merging threshold value.

8. The terminal of claim 6 , wherein, to determining the adaptive target cluster merging threshold value, the processor is further configured to:
acquire a preset cluster merging threshold value;
determine the angular face relationship indicates that the first image and the second image contain face poses of different angles; and
decrease the preset cluster merging threshold value and use the decreased preset cluster merging threshold value as the adaptive target cluster merging threshold value.

9. The terminal of claim 6 , wherein, to determining the adaptive target cluster merging threshold value, the processor is further configured to:
acquire a preset cluster merging threshold value;
determine the angular face relationship indicates that the first image and the second image both contain front view face poses; and
use the preset cluster merging threshold value as the adaptive target cluster merging threshold value.

10. The terminal of claim 6 , wherein, to cluster the first image and the second image, the processor is further configured to:
calculate a cluster merging value;
determine the calculated cluster merging value is greater than the adaptive target cluster merging threshold value; and
determine the first image and the second image belong to a same cluster.

11. A non-transitory computer readable storage medium comprising instructions, wherein the instructions, when executed by a processor in a terminal device, cause the terminal device to:
acquire two images including a first image depicting a first face and a second image depicting a second face;
calculate a similarity between the first face and the second face;
determine the calculated similarity is greater than a preset similarity threshold value:
determine a first face pose information for the first image, the first face pose information including a first orientation value of the first face;
determine a second face pose information for the second image, the second face pose information including a second orientation value of the second face, wherein both the first face pose information and the second face pose information are selected from a pre-established face pose classification mode;
determine, based on the first face pose information and the second face pose information, an angular face relationship between the first face and the second face;
determine a adaptive target cluster merging threshold value based on the angular face relationship; and
cluster the first image and the second image into a same, or different, cluster according to the determined adaptive target cluster merging threshold value.

Assignees

• Xiaomi Inc

Inventors

• Zhang Tao
• Long Fei
• Chen Zhijun

Classifications

• G06V10/763

  Non-hierarchical techniques, e.g. based on statistics of modelling distributions · CPC title

• G06V40/172

  Classification, e.g. identification · CPC title

• G06V40/161 Primary

  Detection; Localisation; Normalisation · CPC title

• G06F18/23211

  with adaptive number of clusters · CPC title

• G06F18/285 Primary

  Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system · CPC title