Picture recognition method and apparatus, electronic device, and medium

Patent metadata

Field	Value
Publication number	US-12567271-B2
Application number	US-202318137884-A
Country	US
Kind code	B2
Filing date	Apr 21, 2023
Priority date	Dec 4, 2020
Publication date	Mar 3, 2026
Grant date	Mar 3, 2026

Abstract

The present disclosure discloses a picture recognition method performed by a computing device. The method includes: obtaining a recognized current string and a hash value of the current string during text recognition for an acquired image through an optical character recognition model, and storing the current string and the hash value of the current string into a first preset tree structure and a second preset tree structure; predicting a new probability value of the current string at a next moment, to obtain an extended string set; obtaining N strings with the highest probability value according to the hash value stored in the second preset tree structure and retaining the N strings; and using the N strings as the current strings, repeating the foregoing steps until recognition of all acquired images is completed, and obtaining a string with the highest probability value as a final recognition result.

First claim

What is claimed is:

1 . A picture recognition method comprising:
at a computing device that has at least one processor and a non-transitory computer-readable medium:
while recognizing text for an acquired image through an optical character recognition model:
obtaining, based on the acquired image, a current string using the optical character recognition model;
computing a hash value of the current string;
storing the current string and the hash value of the current string into a first preset tree structure;
predicting a new probability value of the current string at a next moment;
storing the hash value of the current string into a second preset tree structure;
generating, based on the current string, a target character set using the optical character recognition model at the next moment;
extending the current string through the target character set recognized at the next moment, to obtain an extended string set;
for each extended string in the extended string set:
computing a respective probability value and a respective hash value; and
storing the respective probability value and the respective hash value into the second preset tree structure;
selecting N strings with one or more highest probability values from the first preset tree structure according to the hash values stored in the second preset tree structure, N being an integer not less than 1; and
using the N strings as current strings, repeating the foregoing steps to obtain a target string with a highest probability value as a final recognition result.

2 . The method according to claim 1 , wherein:
the obtaining a current string comprises obtaining the current string formed by characters whose recognition probability is greater than a preset probability; and the computing a hash value of the current string comprises obtaining the hash value of the current string while recognizing text for the acquired image through the optical character recognition model.

3 . The method according to claim 2 , wherein the predicting a new probability value of the current string at a next moment, and storing the hash value of the current string into a second preset tree structure comprises:
predicting a predicted character of the current string at the next moment; obtaining a combined probability of the current string and the predicted character as the new probability value; and storing the hash value of the current string into the second preset tree structure, wherein the second preset tree structure is a set structure.

4 . The method according to claim 1 , wherein the extending the current string through the target character set to obtain an extended string set comprises:
obtaining characters whose recognition probability is greater than a preset probability and recognized at the next moment to form the target character set; and combining the current string with each character in the target character set to obtain all combined strings as the extended string set.

5 . The method according to claim 4 , wherein the selecting N strings with one or more highest probability values from the first preset tree structure according to the hash values stored in the second preset tree structure comprises:
determining N hash values with one or more highest probability values from the second preset tree structure according to the new probability value of the current string and the respective probability value of each extended string; and determining the N strings corresponding to the N hash values from the first preset tree structure according to the N hash values.

6 . The method according to claim 1 , wherein the first preset tree structure is a set structure.

7 . A computing device, comprising a processor and a memory, the memory storing instructions that, when executed by the processor, cause the computing device to perform a picture recognition method including:
while recognizing text for an acquired image through an optical character recognition model:
obtaining, based on the acquired image, a current string using the optical character recognition model;
computing a hash value of the current string;
storing the current string and the hash value of the current string into a first preset tree structure;
predicting a new probability value of the current string at a next moment;
storing the hash value of the current string into a second preset tree structure;
generating, based on the current string, a target character set using the optical character recognition model at the next moment;
extending the current string through the target character set to obtain an extended string set;
for each extended string in the extended string set:
computing a respective probability value and a respective hash value; and
storing the respective probability value and the respective hash value into the second preset tree structure;

selecting N strings with one or more highest probability values from the first preset tree structure according to the hash values stored in the second preset tree structure, N being an integer not less than 1; and
using the N strings as current strings, repeating the foregoing steps to obtain a target string with a highest probability value as a final recognition result.

8 . The computing device according to claim 7 , wherein:
the obtaining a current string comprises obtaining the current string formed by characters whose recognition probability is greater than a preset probability; and the computing a hash value of the current string comprises obtaining the hash value of the current string while recognizing text for the acquired image through the optical character recognition model.

9 . The computing device according to claim 8 , wherein the predicting a new probability value of the current string at a next moment, and storing the hash value of the current string into a second preset tree structure comprises:
predicting a predicted character of the current string at the next moment; obtaining a combined probability of the current string and the predicted character as the new probability value; and storing the hash value of the current string into the second preset tree structure, wherein the second preset tree structure is a set structure.

10 . The computing device according to claim 7 , wherein the extending the current string through the target character set to obtain an extended string set comprises:
obtaining characters whose recognition probability is greater than a preset probability and recognized at the next moment to form the target character set; and combining the current string with each character in the target character set to obtain all combined strings as the extended string set.

11 . The computing device according to claim 10 , wherein the selecting N strings with one or more highest probability values from the first preset tree structure according to the hash values stored in the second preset tree structure comprises:
determining N hash values with one or more highest probability values from the second preset tree structure according to the new probability value of the current string and the respective probability value of each extended string; and determining the N strings corresponding to the N hash values from the first preset tree structure according to the N hash values.

12 . The computing device according to claim 7 , wherein the first preset tree structure is a set structure.

13 . A non-transitory computer-readable storage medium, comprising instructions that, when executed by a processor of a computing device, cause the computing device to perform a picture recognition method including:
while reco

Assignees

• Beijing Sogou Tech Dev Co

Inventors

• Wang Congtao
• Yang Jiajun
• Wei Yuanming
• Chen Wei
• Wei Tao
• Wu Jun
• Gong Li
• Zhu Weiji

Classifications

• H04L9/3236

  using cryptographic hash functions · CPC title

• G06F40/154

  Tree transformation for tree-structured or markup documents, e.g. XSLT, XSL-FO or stylesheets · CPC title

• G06F18/24323

  Tree-organised classifiers · CPC title

• G06F18/2323

  based on graph theory, e.g. minimum spanning trees [MST] or graph cuts · CPC title

• G06V30/262

  using context analysis, e.g. lexical, syntactic or semantic context · CPC title