Medical Image Binarization via Co-occurrence Matrix Thresholding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional thresholding techniques in medical image processing fail to account for the proximity of pixels within foreground and background regions, leading to suboptimal separation of these regions.
Innovation Solution
A method is developed to compute binarization scores based on transition pixel counts using a specified threshold value, which selects a threshold that reduces foreground-to-background and background-to-foreground transitions, thereby increasing the occurrence of adjacent pixels within the same region, using a gray-scale co-occurrence matrix and binary co-occurrence matrix calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional thresholding techniques (e.g., Otsu) are used to separate foreground and background, then the separation is performed based on gray level histogram, but the method fails to account for pixel proximity and leads to suboptimal separation accuracy
Solution Approach 1:
The patent transitions from one-dimensional histogram-based thresholding to two-dimensional analysis by incorporating spatial proximity information through co-occurrence matrices. This adds a spatial dimension to the threshold selection process, allowing simultaneous consideration of gray level distribution and pixel adjacency patterns to achieve more accurate foreground-background separation.
Solution Approach 2:
The patent introduces co-occurrence matrices as an intermediary structure that mediates between the gray level histogram and the threshold selection process. These matrices capture spatial relationships and pixel proximity information, serving as an intermediate representation that enables more informed threshold determination beyond simple gray level statistics.
2Stability of the object's composition
If a binarization threshold is selected to minimize intra-class variance, then the separation is optimized for statistical homogeneity, but transitions between foreground and background increase
Solution Approach 1:
The patent applies local quality by analyzing pixel co-occurrence patterns and spatial relationships in different regions of the image. Instead of treating all pixels uniformly, the method considers local pixel adjacencies and transitions, allowing the threshold selection to account for spatial continuity and reduce unnecessary transitions between foreground and background while maintaining class homogeneity.
Data Source
AI summary
A method for processing an image includes computing a plurality of binarization scores for an image using a corresponding plurality of binarization threshold values, determining a selected binarization threshold value based on the plurality of binarization scores, processing the image or another image using the selected binarization threshold value. Each binarization score of the plurality of binarization scores may be based on a transition pixel count for a binary image that is derivable from the image using a specified threshold value of the plurality of thresholding values. A computer system and computer program product corresponding to the method are also disclosed herein.