Handwriting Recognition Data Augmentation via Style Transfer
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Handwriting recognition systems face challenges due to data distribution shifts caused by varying handwriting styles, leading to concept and prior drift, and suffer from data scarcity and privacy concerns when transferring customer data for training.
Innovation Solution
A data augmentation technique that synthesizes training data in both style and content, using a neural network-based approach to generate augmented data sets, which reduces concept drift and prior drift, and addresses data scarcity and privacy issues by creating artificially generated data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If customer handwriting data is collected and transferred to service provider for training, then training data sufficiency is improved, but data privacy is compromised
Solution Approach 1:
The patent creates synthetic copies of handwriting data through style transfer technology. Instead of using real customer handwriting data, the system generates artificial training data by transferring the style characteristics of customer handwriting onto template characters. This copying approach provides sufficient training data while avoiding the privacy issues of handling and transferring sensitive customer information.
Solution Approach 2:
The patent introduces template characters as an intermediary between the customer's handwriting style and the training data. The style transfer process uses these templates as mediators to generate synthetic handwriting samples. This intermediary approach allows the system to learn from customer handwriting patterns without directly processing or storing sensitive customer data.
2Adaptability or versatility
If transfer learning is applied to address data distribution shift, then model adaptability to new data is improved, but data sufficiency remains insufficient
Solution Approach 1:
The patent performs preliminary data preparation by generating synthetic training data before the actual training process. By pre-generating augmented handwriting samples with diverse styles and conditions, the system ensures sufficient training data is available for transfer learning to be effective. This preliminary action of data synthesis resolves the data sufficiency issue that would otherwise limit transfer learning.
3Reliability
If real customer handwriting data is used for training, then training data relevance is improved, but data privacy risks increase
Solution Approach 1:
The system creates synthetic copies of customer handwriting by transferring style attributes onto template characters. These copied samples maintain the stylistic characteristics and relevance of real customer handwriting while containing no actual sensitive information. This approach preserves training data relevance while eliminating privacy risks associated with using real customer data.
4Quantity of substance
If data augmentation is performed using traditional methods, then data quantity is increased, but data style consistency deteriorates
Solution Approach 1:
The patent changes the parameters of data augmentation by using style transfer technology instead of traditional transformation methods. By controlling the style transfer process with learned parameters from customer handwriting, the system generates augmented data that maintains consistent stylistic characteristics. This parameter-based approach ensures both increased data quantity and preserved style consistency.
Data Source
AI summary
A content aware and style aware neural network based data augmentation model generates augmented data sets to train neural network based handwriting recognition models to recognize individuals' handwriting. The augmented data sets may be generated so as to be artificial, and to lack personal or confidential information. The data augmentation model may generate content reference sets of individual characters generated in different fonts, and style reference sets of pluralities of characters of a particular style, for example, an individual's handwriting.