OCR Hybrid Decoders with Shared Feature Extractor
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current optical character recognition (OCR) systems face challenges in accurately recognizing hand-printed text due to variations in handwriting styles and quality, which can lead to inconsistent results and increased processing requirements.
Innovation Solution
The system employs a hybrid approach using a combination of connectionist temporal classification (CTC) and attention-based decoders, sharing a common feature extractor to generate intermediary outputs, and selects the output with the higher quality metric value based on a language model for improved recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single decoder is used for OCR, then the device complexity is low, but the recognition accuracy is insufficient due to handwriting variations
Solution Approach 1:
The system segments the text recognition task by employing multiple specialized decoders (e.g., CTC decoder for sequence alignment, attention-based decoder for contextual understanding) that each handle specific aspects of handwriting recognition. This segmentation allows each decoder to optimize for its specific function, improving overall recognition accuracy while managing complexity through modular design
Solution Approach 2:
The system merges multiple decoder outputs by combining their intermediary results and selecting the best final output based on quality metrics. This merging approach leverages the complementary strengths of different decoder architectures to achieve higher recognition accuracy than any single decoder could provide alone
2Measurement precision
If multiple decoders are used to improve accuracy, then the recognition quality improves, but the processing time increases
Solution Approach 1:
The system performs preliminary processing by extracting features once and generating intermediary outputs from multiple decoders in parallel before final selection. This preliminary action on feature extraction and intermediary generation allows the system to evaluate multiple hypotheses efficiently without redundant processing, reducing overall processing time while maintaining high recognition quality
Solution Approach 2:
The system applies partial action by selectively processing only the necessary decoder outputs based on quality metric thresholds. Instead of fully processing all possible decoder combinations, the system evaluates intermediary outputs and selects the best one, performing excessive action only when needed to achieve the required accuracy level
3Measurement precision
If multiple decoders generate different hypotheses, then the recognition accuracy improves, but the processing requirements increase
Solution Approach 1:
The system introduces intermediary outputs as intermediate representations that capture essential features from multiple decoders before final decision-making. These intermediary outputs serve as mediators that consolidate information from different decoders, allowing the system to evaluate multiple hypotheses without fully processing each complete hypothesis, thereby reducing computational energy requirements while maintaining improved recognition accuracy
Data Source
AI summary
Embodiments of the present disclosure describe a system and method for optical character recognition. In one embodiment, a system receives an image depicting text. The system extracts features from the image using a feature extractor. The system applies a first decoder to the features to generate a first intermediary output. The system applies a second decoder to the features to generate a second intermediary output, wherein the feature extractor is common to the first decoder and the second decoder. The system determines a first quality metric value for the first intermediary output and a second quality metric value for the second intermediary output based on a language model. Responsive to determining that the first quality metric value is greater than the second quality metric value, the system selects the first intermediary output to represent the text.