Handwritten Character Recognition Block Merging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing OCR technologies face challenges in accurately recognizing handwritten characters that are written discontinuously, as they are typically separated into multiple character blocks due to irregular line changes or shifts in baseline, leading to decreased recognition accuracy when language models follow a reading order like left to right.
Innovation Solution
An information processing apparatus that separates handwritten character areas from document images, extracts character blocks with a common baseline, and combines them based on their position relationships to generate a single continuous character block, improving recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If handwritten characters are separated into multiple character blocks due to irregular line changes or baseline shifts, then the extraction process becomes simpler for each block, but the character recognition accuracy deteriorates because language models cannot properly follow reading order
Solution Approach 1:
The patent merges multiple separated character blocks that belong to the same logical line into a single unified character block. By detecting position relationships between adjacent character blocks and determining they belong to the same line based on spatial proximity and alignment, the system combines them into one block, allowing language models to process the complete character sequence and thereby improving character recognition accuracy while maintaining extraction simplicity
2Measurement precision
If handwritten characters are treated as a single continuous character block, then character recognition accuracy improves, but the complexity of determining which characters belong together increases
Solution Approach 1:
The patent initially segments the handwritten character region into multiple candidate character blocks based on baseline shifts and line changes. Then, it applies a determination process that evaluates position relationships between adjacent blocks to identify which segments belong to the same logical line. This two-stage approach (segmentation followed by intelligent merging) reduces the complexity of determining continuous character blocks compared to treating the entire region as one block, while still achieving high recognition accuracy
Data Source
AI summary
An object is to improve character recognition accuracy of handwritten characters, originally a single continuous character string, described discontinuously. An image area corresponding to a handwritten character is separated from a document image obtained by scanning a document and a character block including characters having the same baseline is extracted. Then, in a case where a plurality of character blocks is extracted from the first image area, a single character block is generated by combining character blocks based on a position relationship of the plurality of character blocks.