Method for recognizing picture, method and apparatus for labelling picture, and storage medium

Patent metadata

Field	Value
Publication number	US-10540563-B2
Application number	US-201515529425-A
Country	US
Kind code	B2
Filing date	Dec 10, 2015
Priority date	Jul 13, 2015
Publication date	Jan 21, 2020
Grant date	Jan 21, 2020

Abstract

A method for recognizing a picture, a method and an apparatus for labelling a picture, and a storage medium. The method for recognizing a picture comprises: acquiring a clothes picture labelled with a positioning key point and a clothes wearing region (S310); extracting, based on the locations of the clothes wearing region and the positioning key point, at least one picture region representing a feature region from the clothes picture (S320); and respectively acquiring feature information representing the feature region based on a region feature recognition model corresponding to the feature region (S330). By labelling the positioning key point, a picture representing a clothes content in the clothes picture is positioned, so that the interference of a background content in the clothes picture is reduced. On this basis, a picture region representing a feature region is extracted from a clothes picture based on a clothes wearing region and a labelled positioning key point, and then feature information on the feature region is acquired based on a corresponding region feature recognition model, thereby enhancing the accuracy and precision of recognizing the feature information on clothes in the clothes picture.

First claim

What is claimed is:

1. A method for recognizing a picture, comprising:
defining a positioning region enclosing a segment of a target object, the positioning region having a predetermined shape and a predetermined size;
defining a point located in the positioning region as a positioning key point corresponding to the positioning region, the positioning key point and the positioning region having a predefined position relationship;
acquiring a target object picture labelled with the positioning key point and a region of the target object,
determining a location of the positioning key point in the target object picture;
determining a region having the predetermined shape, the predetermined size, and the predefined position relationship with the positioning key point in the target object picture, as the positioning region corresponding to the labelled positioning key point based on location of the labelled positioning key point;
extracting the positioning region in the target object picture; and
acquiring feature information representing the positioning region based on a region feature recognition model corresponding to the segment enclosed by the positioning region.

2. The method according to claim 1 , wherein the target object is clothes, and the region of the target object is a clothes wearing region.

3. The method according to claim 2 , wherein the predefined position relationship comprises: for an upper portion of the clothes wearing region, two positioning key points on left and right shoulders, one point at collar, and two points outside left and right lower hems.

4. The method according to claim 3 , wherein the feature information comprises information on a clothes type and/or an attribute of the feature region.

5. The method according to claim 4 , wherein the attribute of the feature region comprises at least one of:
a clothes type, a skirt type, a pants type, a collar type, a sleeve type, a clothes length, a skirt length, a pants length and a sleeve length.

6. The method according to claim 3 , wherein the feature region comprises at least one of:
an upper body region, a lower body region, a whole body region, a collar region, a chest region and a sleeve region.

7. A method for labelling a picture, comprising:
acquiring a target object picture;
acquiring, based on a pre-established target object classification model, information of a region of a target object from the target object picture; and
labelling a positioning key point for the target object picture based on a pre-established target object key point positioning model, wherein the pre-established target object key point positioning model defines a positioning region enclosing a segment of the target object, the positioning region having a predetermined shape and a predetermined size, and defines a point located in the positioning region as the positioning key point corresponding to the positioning region, the positioning key point and the positioning region having a predefined position relationship.

8. The method according to claim 7 , wherein the target object is clothes, and the region of the target object is a clothes wearing region.

9. The method according to claim 8 , wherein the method specifically comprises:
acquiring a clothes picture;
acquiring information of the clothes wearing region in the clothes picture based on a pre-established clothes classification model; and
labelling a first positioning key point for the clothes picture based on a pre-established clothes key point positioning model.

10. The method according to claim 9 , wherein the method further comprises:
acquiring rotation angle information of the clothes in the clothes picture based on information of the labelled first positioning key point;
rotating and correcting the clothes picture based on the rotation angle information; and
labelling a second positioning key point for the rotated and corrected clothes picture based on the clothes key point positioning model.

11. The method according to claim 9 , wherein the clothes wearing region divided by the clothes classification model comprises tops, bottoms, split clothing and jumpsuits.

12. The method according to claim 9 , wherein the clothes classification model is a clothes classification model established based on a convolutional neural network.

13. An apparatus for recognizing a picture, comprising:
a target object picture acquiring module, configured to acquire a target object picture labelled with a positioning key point and a region of the target object, wherein a positioning region enclosing a segment of the target object is defined, the positioning region having a predetermined shape and a predetermined size, and a point located in the positioning region is defined as the positioning key point corresponding to the positioning region, the positioning key point and the positioning region having a predefined position relationship;
a target object region picture extracting module, configured to determine a location of the positioning key point in the target object picture; determine a region having the predetermined shape, the predetermined size, and the predefined position relationship with the positioning key point in the target object picture, as the positioning region corresponding to the labelled positioning key point based on location of the labelled positioning key point and extract the positioning region in the target object picture; and
a target object feature information acquiring module, configured to acquire feature information representing the positioning region based on a region feature recognition model corresponding to the segment enclosed by the positioning region.

14. The apparatus according to claim 13 , wherein the target object is clothes, and the region of the target object is a clothes wearing region.

15. The apparatus according to claim 14 , wherein
the target object picture acquiring module is specifically configured to acquire a clothes picture labelled with a positioning key point and a clothes wearing region;
the target object region picture extracting module is specifically configured to extract, based on locations of the positioning key point and the clothes wearing region, at least one picture region representing a feature region from the clothes picture; and
the target object feature information acquiring module is specifically configured to respectively acquire feature information representing the feature region based on a region feature recognition model corresponding to the feature region.

16. The apparatus according to claim 15 , wherein the feature information comprises information on a clothes type and/or an attribute of the feature region.

17. The apparatus according to claim 16 , wherein the attribute of the feature region comprises at least one of:
a clothes type, a skirt type, a pants type, a collar type, a sleeve type, a clothes length, a skirt length, a pants length and a sleeve length.

18. The apparatus according to claim 15 , wherein the feature region comprises at least one of:
an upper body region, a lower body region, a whole body region, a collar region, a chest region and a sleeve region.

19. An apparatus for labelling a picture, comprising:
a target object picture acquiring module, configured to acquire a target object picture;
a region of the target object acquiring module, configured to acquire information, in the target object picture, of a region of the target object based on a pre-established target object classification model; and
a target object key point l

Assignees

• Baidu online network technology beijing co ltd

Inventors

• Meng Xianglong
• Liu Guoyi
• Han Shumin
• Yan Canxiang
• Du Yuning

Classifications

• G06V10/82

  using neural networks · CPC title

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06V10/255

  Detecting or recognising potential candidate objects based on visual cues, e.g. shapes · CPC title

• G06T7/73 Primary

  using feature-based methods · CPC title

• G06F18/24

  Classification techniques · CPC title