Document Image Shade Removal via Hierarchical Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for removing shade from captured images, particularly those involving large objects, suffer from reduced accuracy due to the influence of the object on shade estimation, leading to low readability and quality in printed images.
Innovation Solution
The method determines shaded regions and specific pixels less affected by shade in a luminance image, generates a shade image by referencing adjacent pixel values, and repeatedly replaces pixel values until the specific pixel region disappears, effectively isolating and removing shade components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the existing method segments the captured image into blocks and estimates shade based on color representative values, then the processing is simple and fast, but the shade estimation accuracy deteriorates when large objects are present in the document
Solution Approach 1:
The patent divides the captured image into multiple blocks and further segments each block into a document region and a non-document region. This hierarchical segmentation allows the system to process the image efficiently while accurately identifying regions affected by large objects, thereby maintaining both processing speed and shade estimation accuracy.
Solution Approach 2:
The patent applies different processing strategies to different regions of the image. Specifically, it identifies non-document regions (large objects) and excludes them from shade estimation, while applying shade correction only to document regions. This local differentiation ensures accurate shade estimation without being influenced by large objects.
2Reliability
If the existing method replaces shade estimated values with neighboring block values, then isolated block issues are corrected, but systematic shade estimation errors across the entire image cannot be eliminated
Solution Approach 1:
The patent employs an iterative feedback mechanism where shade estimation is performed, non-document regions are identified and excluded, shade estimation is re-performed on the remaining document regions, and this process repeats until convergence. This feedback loop ensures that systematic errors are progressively eliminated and the final shade estimation accurately reflects only the document regions.
3Ease of operation
If the existing method processes the entire captured image uniformly, then processing is straightforward, but the presence of large objects significantly degrades the quality of printed images
Solution Approach 1:
The patent performs preliminary identification and exclusion of non-document regions (large objects) before conducting shade estimation. By removing these interfering elements in advance, the subsequent shade estimation and correction processes can proceed uniformly and simply across the document regions, ensuring high printed image quality without complex post-processing.
Data Source
AI summary
An information processing method includes determining, in a luminance image that represents luminance components of an input image obtained by capturing a document image, a shaded region affected by shade and a specific pixel region including specific pixels less affected by the shade through a color of an object in the document image by using the luminance components of pixels of the luminance image, and generating a shade image that represents shade components of the input image by referring to pixel values of pixels adjacent to the specific pixels and repeating replacement processing of replacing a pixel value of each of the specific pixels with a value of the shaded region adjacent to the specific pixel until the specific pixel region disappears.