Integrated Character Recognition and Semantic Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing handwritten character recognition systems face challenges with cursive writing, touching strokes, and confusion in shapes, especially in languages with pictorial character origins like Chinese, due to the lack of mutual relationship consideration in recognition and classification stages, leading to errors and resource-intensive hierarchical searches.
Innovation Solution
A method that integrates recognition and semantic stages into a single step using a disassembled CNN into a feature extractor and a sequential classifier, where the sequential classifier is trained based on context and application domain, and the feature extractor is reusable across multiple domains, allowing for faster development and reduced error correction processing time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If N-gram modeling technique is used for handwritten character recognition, then recognition accuracy can be improved by considering contextual relationships, but computing resource consumption increases due to hierarchical search tree construction and sequential processing requirements
Solution Approach 1:
The patent segments the recognition process into two distinct stages: a recognition stage that extracts features and generates hypotheses, and a semantic stage that performs contextual validation using N-gram models. This segmentation allows the computationally intensive N-gram processing to operate only on candidate hypotheses rather than all possible character combinations, significantly reducing computing resource consumption while maintaining recognition accuracy.
Solution Approach 2:
The patent performs preliminary feature extraction and hypothesis generation in the recognition stage before applying the N-gram semantic model. By pre-processing the input to generate a reduced set of candidate hypotheses, the system avoids the need to build and search hierarchical trees for all possible character combinations, thereby reducing computing resource consumption while preserving the benefits of contextual analysis.
2Productivity
If CNN-based recognition system is used, then recognition speed can be improved by processing single characters independently, but recognition accuracy deteriorates due to ignoring mutual relationships between sequential characters
Solution Approach 1:
The patent merges the advantages of both approaches by combining the fast independent character recognition capability of CNNs with the contextual relationship modeling of N-gram systems. The two-stage architecture integrates the recognition stage (providing speed) and the semantic stage (providing accuracy through contextual validation), allowing the system to achieve both high recognition speed and high accuracy simultaneously.
Solution Approach 2:
The patent introduces an intermediary semantic stage that acts as a bridge between the fast CNN recognition stage and the final output. This intermediary stage uses N-gram models to validate and correct CNN predictions by considering contextual relationships, thereby improving recognition accuracy without significantly impacting the overall recognition speed achieved by the CNN stage.
3Measurement precision
If additional semantic stage is added to correct recognition errors, then recognition accuracy can be improved, but processing time increases due to error correction operations
Solution Approach 1:
The patent applies partial action by having the semantic stage process only the candidate hypotheses generated by the recognition stage rather than performing exhaustive error correction on all possible recognition outcomes. This selective processing approach improves recognition accuracy by validating top candidates against contextual rules while avoiding the time-consuming operation of exploring all possible error correction paths.
Data Source
AI summary
A method for character recognition and semantic for natural language processing comprising extracting a sequence of feature vectors from a sequence of input character images by a convolutional neural network (CNN) feature extractor. The sequence of feature vectors comprises a plurality of feature vectors, each feature vector representing an approximate-match of its corresponding input character in the sequence of input character images. The method further comprises applying a sequential classifier sequentially as a sliding window of a size of a plurality consecutive feature vectors upon the sequence of feature vectors from a first feature vector in the sequence of feature vectors to the last feature vector in the sequence of feature vectors; and recognizing an output character for a targeted feature vector among the applied-upon consecutive feature vectors within the sliding window as it is sliding across the sequence of feature vectors.