Document Region Detection Using Luminance Masking and Edge Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Camera-captured document images often contain clutter and artifacts like non-uniform lighting, glare, and optical distortion, making it challenging to accurately determine the document region-of-interest, which can lead to erroneous results in document processing workflows.
Innovation Solution
A method that involves determining a first candidate document region-of-interest mask through luminance-only image processing, combining it with an edge mask, and refining it based on text features to form a final mask, which is then validated by assessing connected components and contrast measures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If camera-captured document images are used for document processing, then convenience and speed of image acquisition are improved, but image quality and accuracy deteriorate due to clutter, non-uniform lighting, glare, and optical distortion
Solution Approach 1:
The patent segments the image processing task into multiple distinct stages: foreground elimination, non-uniformity reduction, thresholding to create first candidate mask, edge detection to create edge mask, combination of masks, and refinement based on text features. Each stage addresses specific artifacts separately, systematically improving detection accuracy while maintaining efficiency
Solution Approach 2:
The patent introduces intermediate processing steps between raw image capture and final document region determination. Foreground elimination and non-uniformity reduction act as intermediary processes that prepare the image by removing artifacts before the critical thresholding and edge detection operations, thereby improving final detection accuracy
2Productivity
If traditional document processing workflows are applied to camera-captured images, then processing can proceed, but erroneous results occur due to inaccurate document region determination
Solution Approach 1:
The patent performs preliminary actions before the main document processing workflow by accurately determining the document region through enhanced mask generation and validation. The multi-stage processing (foreground elimination, non-uniformity reduction, thresholding, edge detection, mask combination, and text feature-based refinement) prepares the image by precisely identifying the document region, ensuring that subsequent OCR and other processing operations work on the correct area and produce reliable results
Solution Approach 2:
The patent incorporates feedback mechanisms through mask validation using text feature detection and connected component analysis. The system evaluates whether the generated mask accurately represents the document region by checking for expected text features and proper connectivity, allowing iterative refinement of the mask to ensure processing accuracy
Data Source
AI summary
Aspects of the present invention are related to systems and methods for identifying a document region in a non-scanned image. A first candidate document region-of-interest mask may be determined from a luminance-only image by foreground elimination, non-uniformity reduction and thresholding. A second candidate document region-of-interest mask may be formed by subtracting from the first candidate document region-of-interest mask an edge mask resulting from detecting edges in the luminance-only input image. The second candidate document region-of-interest mask may be refined, to form a final document region-of-interest mask, based on text features calculated from the luminance-only input image. The document region-of-interest may be extracted using the final document region-of-interest mask.