Character recognition method and device

Patent metadata

Field	Value
Publication number	US-10872274-B2
Application number	US-201816144219-A
Country	US
Kind code	B2
Filing date	Sep 27, 2018
Priority date	Mar 29, 2016
Publication date	Dec 22, 2020
Grant date	Dec 22, 2020

Abstract

A character recognition method and apparatus are disclosed, relate to the field of image recognition technologies, reduce a calculation time, and increase character recognition efficiency. The method includes: obtaining picture data (110); performing calculation on the picture data by using calculation layers shared by a first classifier and a second classifier, to obtain a first result, where the first classifier is a classifier for recognizing a specific character in the picture data, and the second classifier is a classifier for recognizing whether the picture data is a character picture (120); performing calculation by bringing the first result into remaining calculation layers other than the shared calculation layers in the first classifier, to obtain a first probability corresponding to each character (130); performing calculation by bringing the first result into remaining calculation layers other than the shared calculation layers in the second classifier, to obtain a second probability (140); calculating, based on the first probability and the second probability, a confidence value that the picture data is recognized as each character (150); and outputting a character recognition result based on the confidence (160).

First claim

What is claimed is:

1. A character recognition method, comprising:
training parameter values of calculation layers of a first classifier by using character picture samples, wherein the first classifier comprises a first N calculation layers and a last M calculation layers of calculation layers shared by the first classifier and a second classifier; and
fixing parameters of a first N calculation layers of the second classifier with parameters of the first N calculation layers of the first classifier, and training parameter values of a last L layers of the second classifier by using non-character picture samples and the character picture samples;
obtaining picture data;
performing calculation on the picture data by using the calculation layers shared by the first classifier and the second classifier, to obtain a first result, wherein the first classifier is a classifier for recognizing a specific character in the picture data, and the second classifier is a classifier for recognizing whether the picture data is a character picture;
performing calculation by bringing the first result into remaining calculation layers other than the shared calculation layers in the first classifier, to obtain a first probability corresponding to each character;
performing calculation by bringing the first result into remaining calculation layers other than the shared calculation layers in the second classifier, to obtain a second probability;
calculating, based on the first probability and the second probability, a confidence value that the picture data is recognized as each character; and
outputting a character recognition result based on the confidence value.

2. The method according to claim 1 , wherein the calculation layers shared by the first classifier and the second classifier comprises:
a convolutional layer, or a convolutional layer and at least one fully connected layer.

3. The method according to claim 1 , wherein the character is a digit.

4. The method according to claim 3 , wherein the step of obtaining picture data comprises:
partitioning each piece of picture data from a number area of a picture of an identification card.

5. The method according to claim 4 , wherein the step of calculating, based on the first probability and the second probability, a confidence value that the picture data is recognized as each character comprises:
multiplying a maximum first probability by the second probability, to obtain a confidence value that the picture data is a digit corresponding to the maximum first probability.

6. The method according to claim 5 , wherein the step of outputting a character recognition result based on the confidence value comprises:
selecting a number of digits that have the maximum confidence values, the number being consistent with the number of digits on the identification card, and outputting the digits.

7. The method according to claim 6 , wherein partitioning each piece of picture data from a number area of a picture of an identification card comprises recording an order of the piece of picture data on the identification card.

8. The method according to claim 7 , wherein outputting the digits comprises outputting the digits according to the order of the piece of picture data that each digit corresponds to.

9. A system for character recognition, the system comprising:
one or more processors; and
a memory storing instructions that, when executed by the one or more processors, cause the system to perform:
obtaining picture data, by partitioning each piece of picture data from a number area of a picture of an identification card;
performing calculation on the picture data by using calculation layers shared by a first classifier and a second classifier, to obtain a first result, wherein the first classifier is a classifier for recognizing a specific character in the picture data, and the second classifier is a classifier for recognizing whether the picture data is a character picture;
performing calculation by bringing the first result into remaining calculation layers other than the shared calculation layers in the first classifier, to obtain a first probability corresponding to each character, wherein each character is a digit;
performing calculation by bringing the first result into remaining calculation layers other than the shared calculation layers in the second classifier, to obtain a second probability;
calculating, based on the first probability and the second probability, a confidence value that the picture data is recognized as each character, by multiplying a maximum first probability by the second probability to obtain a confidence value that the picture data is a digit corresponding to the maximum first probability; and
outputting a character recognition result based on the confidence value.

10. The system according to claim 9 , wherein the instructions further cause the system to perform:
training parameter values of each calculation layer of the first classifier by using character picture samples, wherein the first classifier comprises first N calculation layers and last M calculation layers; and
fixing parameters of first N calculation layers of the second classifier with parameters of the first N calculation layers of the first classifier, and training parameter values of last L layers of the second classifier by using non-character picture samples and character picture samples.

11. The system according to claim 9 , wherein the calculation layers shared by the first classifier and the second classifier comprises:
a convolutional layer, or a convolutional layer and at least one fully connected layer.

12. The system according to claim 9 , wherein outputting a character recognition result based on the confidence value comprises:
selecting a number of digits that have the maximum confidence values, the number being consistent with the number of digits on the identification card, and outputting the digits.

13. The system according to claim 12 , wherein partitioning each piece of picture data from a number area of a picture of an identification card comprises recording an order of the piece of picture data on the identification card.

14. The system according to claim 13 , wherein outputting the digits comprises outputting the digits according to the order of the piece of picture data that each digit corresponds to.

15. A system for character recognition, comprising:
one or more processors; and
a memory storing:
a first classifier that, when executed by the one or more processors, causes the system to recognize a specific character in picture data, the first classifier comprising at least two convolutional layers, at least two fully connected layers, and at least one Softmax layer; and
a second classifier that, when executed by the one or more processors, causes the system to recognize whether the picture data is a picture of a character, the second classifier comprising at least two convolutional layers, at least two fully connected layers, and at least one Softmax layer,
wherein the first classifier and the second classifier share at least two convolutional layers and at least one fully connected layer.

Assignees

• Alibaba Group Holding Ltd

Inventors

• Mao Xudong
• Shi Xing
• Chu Wei
• Cheng Mengli
• Zhou Wenmeng

Classifications

• G06V30/147

  Determination of region of interest · CPC title

• G06V30/19173

  Classification techniques · CPC title

• G06V10/82 Primary

  using neural networks · CPC title

• G06F18/217

  Validation; Performance evaluation; Active pattern learning techniques · CPC title

• G06F18/24

  Classification techniques · CPC title