Document Image Color Layer Segmentation for OCR Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing technologies face challenges in accurately performing Optical Character Recognition (OCR) on documents with multiple overlaid information layers, as overlapping objects and text interfere with the OCR process, reducing quality and accuracy.
Innovation Solution
A method and system that divide a document image into separate color layers by classifying and clustering patches within a color space, allowing for effective information extraction from specific layers, including monochromatic and polychromatic segments, to isolate and process each layer independently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional OCR processes the entire document image at once, then the processing is simple and fast, but the accuracy deteriorates when stamps and objects overlap with text
Solution Approach 1:
The patent divides the document image into multiple patches and further segments polychromatic patches into monochromatic segments. This segmentation allows the system to process different color layers separately, improving OCR accuracy by eliminating interference from overlapping stamps and objects, while maintaining manageable processing complexity through automated patch generation and clustering
2Measurement precision
If the system divides the image into multiple color layers, then the OCR accuracy improves by isolating text from overlapping objects, but the processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary actions by automatically dividing the image into patches and clustering them into color layers before OCR processing. This pre-processing separates overlapping elements in advance, allowing the subsequent OCR to operate on cleaned, isolated text regions, thereby improving accuracy while the automation of preliminary steps minimizes time overhead
Solution Approach 2:
The patent changes the parameter of color information by transforming the image into a color space where monochromatic and polychromatic patches can be distinguished. By utilizing color as a separating parameter, the system achieves efficient layer separation without complex geometric analysis, reducing processing time while maintaining high accuracy
3Device complexity
If the system processes only monochromatic patches, then the processing is simplified, but information from polychromatic regions is lost
Solution Approach 1:
The patent segments polychromatic patches into multiple monochromatic segments based on their color composition. This segmentation allows the system to process each segment independently using simple monochromatic processing algorithms, while collectively preserving all information from the original polychromatic regions through the combination of segments
Solution Approach 2:
The patent applies different processing qualities to different regions: monochromatic patches are processed as single-color elements, while polychromatic patches are segmented into multiple monochromatic segments. This local differentiation maintains processing simplicity for uniform regions while ensuring information completeness in multi-color regions through segmentation
Data Source
AI summary
A method of the disclosure includes receiving, by a processing device, a document image, dividing the document image into a plurality of patches and determining, for each patch, whether the patch is monochromatic or polychromatic. It further includes clusterizing a plurality of monochromatic patches into a plurality of clusters within a color space, wherein each cluster corresponds to a color layer of a plurality of color layers of the document image, and segmenting each polychromatic patch into a corresponding plurality of monochromatic segments. The method also includes, for each polychromatic patch, associating each monochromatic segment of the corresponding plurality of monochromatic segments with a cluster of the plurality of clusters, and utilizing the plurality of clusters for performing an information extraction task on the document image.