Handwriting Stroke Sorting for Multi-Direction Text Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing online handwriting recognition systems struggle to handle multiple writing directions, particularly for languages like Chinese, Japanese, and Korean, and fail to recognize strokes with reversed orders, while existing optical text recognition methods face difficulties with non-horizontal and non-vertical writing directions.
Innovation Solution
A neural network-based handwriting text recognition apparatus that supports multiple writing directions by calculating the writing direction of strokes using center of gravity and second central moments, sorting strokes based on direction, and employing a Sequence to Sequence model for accurate recognition, eliminating the need for separate data collection for vertical handwriting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a classifier is used to distinguish horizontal and vertical writing directions and process them separately, then horizontal and vertical writing directions can be recognized, but the method encounters difficulties when processing other writing directions and requires complex rotation operations
Solution Approach 1:
The patent applies a unified stroke sorting mechanism that works for all writing directions (horizontal, vertical, and others) without requiring separate processing paths. The sorting operator uses a general angle-based classification system that automatically adapts to any writing direction, eliminating the need for direction-specific handling and reducing overall system complexity.
Solution Approach 2:
The patent changes the parameter used for stroke sorting from fixed directional assumptions to a dynamic angle-based system. By calculating the writing direction angle and using it to determine sorting order, the system can handle any writing direction without requiring complex rotation operations or separate classifiers for each direction.
2Measurement precision
If separate data collection is performed for vertical handwriting, then vertical writing direction recognition accuracy can be improved, but the cost of data collection and labeling increases
Solution Approach 1:
The patent trains a single recognition model that handles multiple writing directions (horizontal, vertical, and others) without requiring separate data collection for each direction. The unified model architecture and angle-based sorting mechanism enable the system to generalize across all writing directions, eliminating the need for separate vertical handwriting data collection and labeling efforts.
3Device complexity
If strokes are processed in the order they are written, then the recognition system is simple, but it cannot handle reversed stroke orders in the same character
Solution Approach 1:
The patent performs preliminary stroke sorting based on writing direction angle before feeding strokes to the recognition model. This pre-sorting action organizes strokes into the correct logical sequence regardless of their actual writing order, enabling the recognition system to handle reversed stroke orders without adding complexity to the core recognition algorithm.
Data Source
AI summary
A handwriting text recognition apparatus supporting multiple writing directions comprising a processor and a memory; a stroke input processor serving to receive a plurality of online handwriting character string tracks which are exploded into a plurality of strokes; a writing direction determination operator connected to the stroke input processor for calculating a writing direction of the online handwriting character string tracks by using a coordinate of a center of gravity and a plurality of second central moments of a plurality of sampling points of the strokes; a stroke sorting operator connected to the writing direction determination operator and sorting the strokes by the writing direction to obtain an x direction labeled sorting stroke set or a y direction labeled sorting stroke set; and wherein the x direction labeled sorting stroke set or the y direction labeled sorting stroke set is input to a recognition model for recognition.