Document Image Edge Reconstruction and Perspective Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image processing methods struggle to accurately detect and correct faulty images, especially when background color and texture match, leading to inaccuracies in identifying missing edges and corners, which hinders automatic information retrieval from documents like identification cards.
Innovation Solution
An image processing system that detects edges, reconstructs missing edges by cloning pixels based on color gradients, identifies outer corner points using quadratic corner points, and performs perspective correction to produce a processed image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image processing methods are used to detect edges and corners, then the processing can be performed quickly, but the accuracy deteriorates when background color and texture match the captured image
Solution Approach 1:
The patent converts the captured image from RGB color space to LAB color space, which separates luminance (L) from color information (A and B channels). This transformation enhances the contrast between the document edges and background by operating on the A and B channels independently, allowing edge detection to succeed even when the background color closely matches the document color in the original RGB space.
Solution Approach 2:
The patent applies morphological operations (dilation and erosion) with varying kernel sizes and iteratively adjusts detection parameters to enhance edge visibility. By changing the structural parameters of the morphological operations and iterating through multiple detection passes, the system adapts to different image conditions and overcomes background interference that would prevent accurate edge detection using fixed parameters.
2Reliability
If manual verification is performed to retrieve information from faulty images, then the accuracy of information retrieval is maintained, but the time consumption increases significantly
Solution Approach 1:
The patent performs preliminary image correction operations including perspective transformation, skew correction, and distortion correction before the information retrieval process. By pre-correcting the image geometry and alignment issues, the system ensures that subsequent automated OCR and information extraction can proceed accurately without requiring manual verification, thus maintaining reliability while reducing time loss.
Solution Approach 2:
The system implements automated self-correction of image defects by detecting and correcting perspective distortions, skew angles, and geometric transformations using feature point detection and homography calculations. This self-service capability allows the system to automatically compensate for poor image quality and alignment without human intervention, eliminating the need for manual verification in many cases.
3Measurement precision
If image perspective correction is performed to correct skew and distortion, then the accuracy of information retrieval is improved, but the device complexity increases
Solution Approach 1:
The patent divides the perspective correction process into distinct modular stages: (1) edge detection and corner point identification, (2) homography matrix calculation from four corner points, (3) perspective transformation application, and (4) skew/distortion correction. By segmenting the complex correction process into independent modules, the system achieves high alignment accuracy while keeping each individual module relatively simple and manageable.
Data Source
Figure 1
Figure 2
Figure 3a
AI summary
The present disclosure relates to a method and system for processing an image extracted from a document. The image processing system detects one or more edges of the image and determines a missing edge in the image by comparing color gradients of the edges. The missing edges are recreated by cloning image pixels based on data pixels of an edge, opposite to the missing edge, amongst the one or more edges to obtain a reconstructed image. Outer corner points in the reconstructed image are identified based on quadratic corner points, wherein the quadratic corner points are determined based on the one or more edges and the reconstructed edge. Further, the image processing system performs an image perspective correction on the reconstructed image, based on the outer corner points, to obtain a processed image.