Handwriting Recognition Using Lexical Knowledge Base
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current handwriting recognition techniques face challenges in efficiency, reliability, and cost-effectiveness due to sensitivity to segmentation errors, requirement for large training sets, and difficulty in modeling variability in stroke sequences, especially in off-line recognition systems without dynamic information.
Innovation Solution
A process that decomposes cursive traces into image fragments, classifies them as isolated characters or portions of cursive writing, segments them into strokes, and uses a dynamic Lexicon and Reference Set to validate interpretations, reducing the need for extensive training and feature extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If analytical techniques are used to recognize single characters, then the system can recognize any sequence of characters, but the system becomes extremely sensitive to segmentation errors and requires very numerous training sets
Solution Approach 1:
The patent introduces a lexical knowledge base as an intermediary between the segmented character fragments and the final recognition result. The system segments the handwritten trace into fragments, then uses the lexical knowledge base containing valid words and language rules to validate and select the correct interpretation, thereby reducing sensitivity to segmentation errors
Solution Approach 2:
The system performs preliminary organization of linguistic knowledge into a structured lexical knowledge base before the recognition process. This pre-organized knowledge structure enables efficient validation and disambiguation during recognition, reducing the need for numerous training sets
2Ease of operation
If holistic techniques are used to recognize whole words, then segmentation is not required, but the system needs as many different classes as the number of different words and requires very large training sets
Solution Approach 1:
The patent applies segmentation at the fragment level rather than requiring full-word segmentation. By dividing the handwritten trace into smaller fragments and using a lexical knowledge base to reconstruct valid words from these fragments, the system achieves ease of operation without the complexity of holistic techniques
Solution Approach 2:
The system changes the parameter of class organization from word-level classes to character/fragment-level classes combined with lexical constraints. This parameter change reduces the number of required classes while maintaining recognition capability through the lexical knowledge base
3Adaptability or versatility
If stroke-based methods are used in off-line recognition, then the system can model variability in stroke sequences, but extracting actual strokes becomes much harder without dynamic information
Solution Approach 1:
The patent uses inexpensive image processing techniques to extract stroke-like features from static images without requiring complex dynamic information. The system processes the image data directly using affordable computational methods, avoiding the need for expensive or complex stroke extraction algorithms
Solution Approach 2:
The lexical knowledge base serves as an intermediary that validates stroke sequences without requiring precise extraction of actual strokes. The system can work with approximate stroke representations and use the lexical knowledge to confirm valid interpretations, reducing the difficulty of stroke detection
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Process, and related apparatus, that exploits psycho-physiological aspects involved in generation and perception of handwriting for directly inferring from the trace on the paper (or any other means on which the author writes by hand) the interpretation of writing, i.e. the sequence of characters that the trace is intended to represent.