Method and device for recognizing image and storage medium

Patent metadata

Field	Value
Publication number	US-2020320352-A1
Application number	US-202016910716-A
Country	US
Kind code	A1
Filing date	Jun 24, 2020
Priority date	Jun 24, 2019
Publication date	Oct 8, 2020
Grant date	—

Abstract

A method and device for recognizing an image, electronic equipment and a storage medium are provided. The method includes: acquiring an image to be recognized; determining a potential recognition region based on a target algorithm model; determining an up-sampled potential recognition region by up-sampling the potential recognition region; and determining a classification recognition result based on the up-sampled potential recognition region.

First claim

What is claimed is:

1 . A method for recognizing an image, comprising:
acquiring an image to be recognized; determining a potential recognition region of the image based on a target algorithm model, wherein the potential recognition region includes a region with a designated content and a size no greater than a preset threshold; determining an up-sampled potential recognition region by up-sampling the potential recognition region; and determining a classification recognition result based on the up-sampled potential recognition region.

2 . The method of claim 1 , wherein the target algorithm model comprises a feature extraction network, a region proposal network and a region pooling network.

3 . The method of claim 2 , wherein said determining the potential recognition region comprises:
determining a first feature map by extracting features from the image based on the feature extraction network; determining a first predict bounding box based on the first feature map and the region proposal network, wherein the first predict bounding box includes a target feature region; determining a target recognition result based on the region pooling network and the target feature region; determining the potential recognition region based on the target recognition result.

4 . The method of claim 3 , wherein the target recognition result comprises:
the target feature region includes no potential recognition region; the target feature region includes the potential recognition region.

5 . The method of claim 4 , wherein the target recognition result comprises a position of the potential recognition region and a classification recognition result of regions other than the potential recognition region in the target feature region in response to that the target feature region includes the potential recognition region.

6 . The method of claim 4 , wherein said determining the potential recognition region based on the target recognition result comprises:
extracting the potential recognition region from the image in response to that the target feature region includes the potential recognition region.

7 . The method of claim 2 , wherein said determining the classification recognition result comprises:
determining a second feature map by extracting features from the up-sampled potential recognition region based on the feature extraction network; determining a second predict bounding box based on the second feature map and the region proposal network, wherein the second predict bounding box includes a specified feature region; and determining the classification recognition result based on the region pooling network and the specified feature region.

8 . The method of claim 1 , wherein the target algorithm model is pre-trained by:
acquiring sample images; determining labeled sample images by labeling designated contents and potential recognition regions in the sample images; and obtaining the target algorithm model by training an initial algorithm model based on the labeled sample images.

9 . A device for recognizing an image, comprising:
a processor; and a memory configured to store instructions executable by the processor; wherein the processor is configured to execute the instructions to: acquire an image to be recognized; determine a potential recognition region of the image based on a target algorithm model, wherein the potential recognition region includes a region with a designated content and a size no greater than a preset threshold; determine an up-sampled potential recognition region by up-sampling the potential recognition region; and determine a classification recognition result based on the up-sampled potential recognition region.

10 . The device of claim 9 , wherein the target algorithm model comprises a feature extraction network, a region proposal network and a region pooling network.

11 . The device of claim 10 , wherein the processor is configured to:
determine a first feature map by extracting features from the image based on the feature extraction network; determine a first predict bounding box based on the first feature map and the region proposal network, wherein the first predict bounding box includes a target feature region; determine a target recognition result based on the region pooling network and the target feature region; determine the potential recognition region based on the target recognition result.

12 . The device of claim 11 , wherein the target recognition result comprises:
the target feature region includes no potential recognition region; the target feature region includes the potential recognition region.

13 . The device of claim 12 , the target recognition result comprises a position of the potential recognition region and a classification recognition result of regions other than the potential recognition region in the target feature region in response to that the target feature region includes the potential recognition region.

14 . The device of claim 12 , wherein the processor is configured to:
extract the potential recognition region from the image in response to that the target feature region includes the potential recognition region.

15 . The device of claim 10 , wherein the processor is configured to:
determine a second feature map by extracting features from the up-sampled potential recognition region based on the feature extraction network; determine a second predict bounding box based on the second feature map and the region proposal network, wherein the second predict bounding box includes a specified feature region; and determine the classification recognition result based on the region pooling network and the specified feature region.

16 . The device of claim 9 , wherein the target algorithm model is pre-trained by:
acquiring sample images; determining labeled sample images by labeling the designated contents and potential recognition regions in the sample images; and obtaining the target algorithm model by training an initial algorithm model based on the labeled sample images.

17 . A computer readable storage medium storing computer programs that, when executed by a processor, cause the processor to perform the operation of:
acquiring an image to be recognized; determining a potential recognition region of the image based on a target algorithm model, wherein the potential recognition region includes a region with a designated content and a size no greater than a preset threshold; determining an up-sampled potential recognition region by up-sampling the potential recognition region; and determining a classification recognition result based on the up-sampled potential recognition region.

Assignees

• Beijing Dajia Internet Information Tech Co Ltd

Inventors

• Zhang Shuifa
• Li Yan
• Wang Sibo
• Liu Chang

Classifications

• G06V30/147 Primary

  Determination of region of interest · CPC title

• G06V30/10

  Character recognition · CPC title

• G06V40/172

  Classification, e.g. identification · CPC title

• G06F18/241

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

• G06K9/6268 Primary

  Physics · mapped topic