Superimposed Handwriting Recognition via Segmentation Graph
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing on-line handwriting recognition systems for computing devices face challenges in accurately interpreting natural handwriting with multi-stroke characters on small input surfaces, where characters are superimposed, leading to difficulties in determining character boundaries and clear display of input characters.
Innovation Solution
A system and method for on-line superimposed handwriting recognition that uses a segmentation expert to detect relative positions and geometry of input strokes, classifies fragments, and employs a recognition engine to generate a segmentation graph with nodes for character hypotheses, assigning recognition scores and generating linguistic meaning to analyze input strokes simultaneously.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If multi-stroke recognition is used to recognize natural handwriting, then ease of operation is improved, but difficulty of detecting and measuring increases due to superimposed characters on small input surfaces
Solution Approach 1:
The system segments the superimposed handwriting input into multiple temporal fragments or strokes, analyzing each segment's geometric features and relative positions. This segmentation allows the system to process complex multi-stroke characters by breaking them down into manageable components that can be individually characterized and then recombined to identify complete characters and their boundaries.
Solution Approach 2:
The system adds a temporal dimension to the analysis by treating superimposed handwriting as a sequence of strokes occurring at different times. By analyzing the temporal order and relative positions of strokes across multiple fragments, the system can distinguish character boundaries even when characters are spatially overlapping, effectively moving from a 2D spatial problem to a 3D space-time problem.
2Measurement precision
If explicit character separation is required for accurate recognition, then measurement precision is improved, but ease of operation deteriorates due to additional user input requirements
Solution Approach 1:
The system automatically detects character boundaries by analyzing the geometric features and relative positions of strokes within the superimposed input, without requiring the user to provide explicit separation signals. The algorithm self-determines where one character ends and another begins based on the spatial and temporal relationships between strokes, making the process transparent to the user while maintaining high precision.
Solution Approach 2:
The system uses feedback from the geometric analysis of stroke patterns to iteratively refine character boundary detection. By evaluating the relative positions and orientations of successive strokes, the system can identify characteristic patterns that indicate character boundaries, adjusting its segmentation hypotheses based on the accumulated geometric evidence from multiple strokes.
3Reliability
If the number of character hypotheses is increased to cover all possible interpretations, then reliability is improved, but processing time increases due to simultaneous analysis requirements
Solution Approach 1:
The system generates a limited set of plausible character hypotheses based on the geometric features of the input strokes, rather than exhaustively considering all possible interpretations. By focusing computational resources on the most likely character candidates that match the observed stroke patterns, the system achieves high reliability without the prohibitive processing time that would result from analyzing every possible hypothesis.
Solution Approach 2:
The system dynamically adjusts recognition parameters such as character probability thresholds and hypothesis generation criteria based on the specific geometric characteristics of the input. By adapting these parameters to the observed stroke patterns, the system can maintain high recognition accuracy while controlling the number of hypotheses that require full processing, thereby optimizing the balance between reliability and processing speed.
Data Source
Figure 1
Figure 2
Figure 3A~3B
AI summary
A system and method that is able to recognize a user's natural superimposed handwriting without any explicit separation between characters. The system and method is able to process single- stroke and multi-stroke characters. It can also process cursive handwriting. Further, the method and system can determine the boundaries of input words either by the use of a specific user input gesture or by detecting the word boundaries based on language characteristics and properties. The system and method analyzes the handwriting input through the processes of segmentation (116), character recognition (118), and language modeling (120). At least some of these processes occur concurrently through the use of dynamic programming.