Two-Pass OCR Re-Zoning for Small Text Zone Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing OCR systems have difficulty detecting small text zones such as headers, footers, and page numbers, leading to inaccurate character recognition and limited improvement from post-OCR spell checking due to lack of surrounding context.
Innovation Solution
A two-pass OCR method where the first pass identifies initial text zones, and a re-zoning engine analyzes the output to identify new text zones based on patterns, such as headers, footers, and incremental sequences, with a second pass performing character recognition on these newly identified zones using textual pattern templates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single-pass OCR method is used, then the processing speed is fast, but the recognition accuracy is limited due to missed text zones
Solution Approach 1:
The OCR process is divided into two distinct passes: a first pass that performs initial character recognition on identified text zones, and a second pass that performs additional recognition on newly identified text zones using pattern templates. This segmentation allows the system to achieve higher accuracy by processing different zones in separate stages rather than attempting to handle all zones in a single complex pass.
Solution Approach 2:
The first pass of OCR is performed as a preliminary action to identify initial text zones and generate initial textual content. This preliminary recognition provides a foundation for the second pass, which uses pattern templates to identify additional text zones that were missed in the first pass. The preliminary action enables the system to build upon initial results rather than starting from scratch.
2Measurement precision
If dictionary-based spell correction is used, then some recognition errors are corrected, but new errors are introduced and comprehensiveness is limited
Solution Approach 1:
Pattern templates are created in advance based on common document structures (headers, footers, page numbers, captions). These templates serve as preliminary recognition rules that guide the second pass OCR process, enabling the system to identify text zones based on structural patterns rather than relying solely on dictionary-based spell correction. This preliminary structuring allows the system to handle diverse document types without requiring an exhaustive dictionary.
3Measurement precision
If traditional single-pass OCR is used, then processing is efficient, but small text zones like headers and footers are missed
Solution Approach 1:
The document processing is segmented into two passes: the first pass handles general text zones efficiently, while the second pass specifically targets newly identified text zones using pattern templates. This segmentation allows the system to focus computational resources on areas where additional recognition is most beneficial, rather than uniformly processing the entire document twice.
Solution Approach 2:
The second pass performs OCR only on newly identified text zones rather than re-processing the entire document. This partial action approach applies the additional processing effort only where needed - specifically on zones that were missed or not confidently identified in the first pass - thereby reducing overall processing time while still improving detection accuracy for small text zones.
Data Source
AI summary
An image of a paginated document is zoned to identify text zones. First-pass character recognition is performed on the text zones to generate textual content corresponding to the paginated document. The image of the paginated document is re-zoned based on the textual content to identify one or more new text zones. Second-pass character recognition is performed on at least the new text zones to generate updated textual content corresponding to the paginated document.