Handwriting Recognition System Using Online Offline Score Combination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing handwritten input recognition models, whether online or offline, suffer from recognition accuracy issues due to different error sets and dataset dependencies, leading to suboptimal performance.
Innovation Solution
Combining online and offline recognition models using statistical analysis, AdaBoost, neural networks, or graph-based methods to integrate scores and produce a unified recognition result, enhancing accuracy by leveraging the strengths of both approaches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If online recognition model is used, then recognition speed is improved, but recognition accuracy deteriorates due to different error sets
Solution Approach 1:
The patent combines online and offline recognition models into a unified system where both models process the same handwritten input and their results are integrated. The online model provides fast initial recognition while the offline model provides accurate verification, resolving the contradiction between speed and accuracy by merging their complementary strengths.
Solution Approach 2:
The recognition system uses a composite approach by integrating multiple recognition models (online HMM model and offline statistical template-based model) with different error characteristics. This composite system leverages the strengths of each model type to achieve both speed and accuracy that neither model could achieve alone.
2Measurement precision
If offline recognition model is used, then recognition accuracy is improved, but processing time increases
Solution Approach 1:
The online recognition model performs preliminary recognition quickly to provide an initial result and candidate hypotheses. This preliminary action filters out obviously incorrect recognitions before the more time-consuming offline model processes the input, reducing the overall processing time while maintaining high accuracy.
Solution Approach 2:
Instead of always applying the full offline recognition process, the system uses the online model to quickly eliminate clearly incorrect candidates, applying the more computationally intensive offline processing only where needed. This partial application of the offline model reduces processing time while preserving accuracy benefits.
3Device complexity
If single recognition model is used, then system complexity is reduced, but recognition accuracy deteriorates due to dataset dependencies
Solution Approach 1:
The system changes the parameter of model diversity by incorporating multiple recognition models with different architectural parameters and error characteristics. This allows the system to adapt to different datasets and handwriting styles, improving accuracy without requiring a completely redesign of the recognition architecture.
Solution Approach 2:
The combined recognition system achieves universality by being able to handle various handwriting styles, datasets, and input types through the complementary strengths of multiple models. The system can adapt to different recognition scenarios while maintaining a unified architecture, balancing complexity and accuracy.
Data Source
AI summary
Described is a technology by which online recognition of handwritten input data is combined with offline recognition and processing to obtain a combined recognition result. In general, the combination improves overall recognition accuracy. In one aspect, online and offline recognition is separately performed to obtain online and offline character-level recognition scores for candidates (hypotheses). A statistical analysis-based combination algorithm, an AdaBoost algorithm, and/or a neural network-based combination may determine a combination function to combine the scores to produce a result set of one or more results. Online and offline radical-level recognition may be performed. For example, a HMM recognizer may generate online radical scores used to build a radical graph, which is then rescored using the offline radical recognition scores. Paths in the rescored graph are then searched to provide the combined recognition result, e.g., corresponding to the path with the highest score.