Method of generating font database, and method of training neural network model

Patent metadata

Field	Value
Publication number	US-11816908-B2
Application number	US-202217683514-A
Country	US
Kind code	B2
Filing date	Mar 1, 2022
Priority date	Apr 20, 2021
Publication date	Nov 14, 2023
Grant date	Nov 14, 2023

Abstract

A method of generating a font database, and a method of training a neural network model are provided, which relate to a field of artificial intelligence, in particular to a computer vision and deep learning technology. The method of generating the font database includes: determining, by using a trained similarity comparison model, a basic font database most similar to handwriting font data of a target user in a plurality of basic font databases as a candidate font database; and adjusting, by using a trained basic font database model for generating the candidate font database, the handwriting font data of the target user, so as to obtain a target font database for the target user.

First claim

What is claimed is:

1. A method of generating a font database, comprising:
determining, by using a trained similarity comparison model, a basic font database most similar to handwriting font data of a target user in a plurality of basic font databases as a candidate font database; and
adjusting, by using a trained basic font database model for generating the candidate font database, the handwriting font data of the target user, so as to obtain a target font database for the target user, wherein
the adjusting, by using the trained basic font database model for generating the candidate font database, the handwriting font data of the target user includes:
using a basic stroke of a standard font as an input of the trained basic font database model for generating the candidate font database; and
using a basic stroke of the handwriting font of the target user as an output of the trained basic font database model for generating the candidate font database; and

the basic stroke of the handwriting font of the target user is obtained by segmenting a handwritten word of the target user using a coherent point drift (CPD) matching algorithm.

2. The method according to claim 1 , wherein the determining a basic font database most similar to handwriting font data of a target user in a plurality of basic font databases as a candidate font database comprises:
determining, by using the trained similarity comparison model, a similarity between the handwriting font data of the target user and corresponding font data in each basic font database of the plurality of basic font databases; and
determining a basic font database with a greatest similarity in the plurality of basic font databases as the candidate font database.

3. The method according to claim 2 , wherein the similarity comprises a sum of similarities between a plurality of handwriting font data of the target user and a plurality of corresponding font data in the each basic font database of the plurality of basic font databases.

4. The method according to claim 1 , further comprising replacing radical data in the target font database with corresponding radical data of the handwriting font data of the target user.

5. The method according to claim 4 , wherein the replacing radical data in the target font database with corresponding radical data of the handwriting font data of the target user comprises:
calculating a difference between a position of a centroid of radical image data after the replacing and a position of a centroid of radical image data before the replacing; and
adjusting, based on the difference, a position of the radical image data after the replacing.

6. The method according to claim 1 , further comprising filtering a matching result obtained using the CPD matching algorithm, so as to remove a result determined to be incorrect.

7. An electronic device, comprising:
at least one processor; and
a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions, when executed by the at least one processor, cause the at least one processor to implement the method of claim 1 .

8. The electronic device according to claim 7 , wherein the at least one processor is further caused to:
determine, by using the trained similarity comparison model, a similarity between the handwriting font data of the target user and corresponding font data in each basic font database of the plurality of basic font databases; and
determine a basic font database with a greatest similarity in the plurality of basic font databases as the candidate font database.

9. The electronic device according to claim 8 , wherein the similarity comprises a sum of similarities between a plurality of handwriting font data of the target user and a plurality of corresponding font data in the each basic font database of the plurality of basic font databases.

10. A non-transitory computer-readable storage medium having computer instructions stored thereon, wherein the computer instructions allow a computer to implement the method of claim 1 .

11. A method of training a neural network model comprising a basic font database model and a similarity comparison model, comprising:
training the basic font database model using handwriting font data of a plurality of users, so that each user of the plurality of users has a corresponding basic font database model and a corresponding basic font database; and
training the similarity comparison model using a plurality of basic font databases of the plurality of users, wherein
the training the basic font database model using handwriting font data of a plurality of users includes:
using a basic stroke of a standard font as an input; and
using a basic stroke of a handwriting font of each user of the plurality of users as an output; and

the basic stroke of the handwriting font of each user of the plurality of users is obtained by segmenting a handwritten word of the user using a coherent point drift (CPD) matching algorithm.

12. The method according to claim 11 , wherein the basic font data comprises font image data, and the training the similarity comparison model using a plurality of basic font databases of the plurality of users comprises:
selecting two image data randomly from N basic font databases in the plurality of basic font databases to form an image pair, where the N basic font databases are represented by an array {X i }, (i=1, 2, . . . , N), and N is a natural number greater than 1;
adding a label to the image data pair, wherein a value of 1 is assigned to the label in response to determining that the image data pair is formed by two image data selected from the same basic font database, and a value of 0 is assigned to the label in response to determining that the image data pair is formed by two image data selected from different basic font databases;
inputting the two image data into a weight sharing network successively to obtain a feature vector v 1 and a feature vector v 2 respectively representing image features of the two image data;
calculating a distance between the feature vector v 1 and the feature vector v 2 ;
determining a loss function for the similarity comparison model according to the distance and the label; and
updating the similarity comparison model using the loss function.

13. The method according to claim 12 , wherein the loss function Loss(i 1 , i 2 , label) for the similarity comparison model is expressed as
Loss( i 1 ,i 2 ,label)=(1−label)*(1−metric( v 1 ,v 2 ))+label*metric( v 1 ,v 2 )

where i 1 and i 2 represent the two image data, respectively, and metric represents a Euclidean distance.

14. The method according to claim 11 , further comprising filtering a matching result obtained using the CPD matching algorithm, so as to remove a result determined to be incorrect.

15. An electronic device, comprising:
at least one processor; and
a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions, when executed by the at least one processor, cause the at least one processor to implement the method of claim 11 .

16. A non-transitory computer-readable storage medium having computer instructions stored thereon, wherein the computer instructions allow a computer to implement the method of claim 11 .

Assignees

• Beijing Baidu Netcom Sci & Tech Co Ltd

Inventors

• Tang Licheng
• Liu Jiaming

Classifications

• G06V30/245 Primary

  Font recognition · CPC title

• G06V10/761

  Proximity, similarity or dissimilarity measures · CPC title

• G06V10/7747

  Organisation of the process, e.g. bagging or boosting · CPC title

• G06V10/82

  using neural networks · CPC title

• G06V30/2455

  Discrimination between machine-print, hand-print and cursive writing · CPC title