Heterogeneous OCR Engine Combination for Text Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional OCR systems face challenges in accurately recognizing text, especially in low-quality images with variations in lighting, shadows, contrast, and blur, leading to imprecision in text detection and recognition.
Innovation Solution
The use of multiple OCR engines running in a multithreaded mode, including a width-focused and a depth-focused recognition engine, which analyze image data concurrently to determine base-sets and focused-sets of words, merge overlapping bounding boxes based on confidence values, and normalize confidence values for improved text recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single conventional OCR engine is used, then the system is simple and resource-efficient, but text recognition precision deteriorates in low-quality images
Solution Approach 1:
The system segments the text recognition task by employing multiple specialized OCR engines (e.g., width-focused and depth-focused engines) that each excel at detecting text in specific orientations or conditions. This segmentation allows the system to maintain high precision across diverse image qualities without requiring a single overly complex engine.
Solution Approach 2:
The system merges the outputs of multiple OCR engines through a combination module that integrates their respective strengths. By combining results from engines specialized in different detection approaches, the system achieves superior overall precision while managing complexity through modular architecture.
2Measurement precision
If multiple OCR engines are used, then text recognition precision improves, but resource intensity increases
Solution Approach 1:
The system applies partial action by having different OCR engines process different aspects or portions of the image data according to their specialized strengths. Rather than all engines processing all data fully, each engine performs targeted analysis, reducing redundant computational effort while maintaining high detection accuracy.
Solution Approach 2:
The system uses simplified copies or specialized versions of OCR engines tailored for specific detection tasks (e.g., horizontal text detection, vertical text detection). These specialized copies require fewer computational resources than a full-featured engine would need to handle all cases, thereby reducing overall resource intensity.
3Reliability
If multiple OCR engines process images concurrently, then text detection capability in complex backgrounds improves, but processing time increases
Solution Approach 1:
The system segments the processing workload by assigning different image regions or text orientation ranges to specialized engines. This allows concurrent processing of divided tasks, improving reliability through multiple perspectives while reducing total processing time compared to sequential execution.
Solution Approach 2:
The system employs periodic or iterative processing where engines take turns analyzing different aspects of the image data. This periodic action allows efficient utilization of computational resources while maintaining comprehensive text detection capability across complex backgrounds.
Data Source
AI summary
Approaches provide for recognizing and locating text represented in image data. For example, image data that includes representations of text can be obtained. A width-focused recognition engine can be configured to analyze the image data to determine a base-set of words. The base-set of words can be associated with logical structure information that describes a geometric relationship between words in the base-set of words. A set of bounding boxes that includes one or more base words can be determined, as well as a confidence value for each base word. A depth-focused recognition engine can be configured to analyze the image data to determine a focused-set of words, the focused-set of words associated with a set of bounding boxes and confidence values for respective words. A set of merged words can be determined from a set of overlapping bounding boxes that overlap a threshold amount. The set of merged words can include at least a portion of the base-set of words and/or the focused-set of words and are selected based at least in part on respective confidence values of words in the set of overlapping bounding boxes. Thereafter, a final set of words that includes the merged set of words and appended words can be determined.