Edge Detection in Noisy Multi-Contrast Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems face challenges in accurately extracting structural information from multi-contrast or multi-modal images due to noise amplification, particularly in spectral computed tomography and differential phase contrast imaging, which hinders effective segmentation of structures or organs.
Innovation Solution
An image processing system that receives multiple input images encoding different physical properties, uses a differentiator to form differences between image pairs, and an edge evaluator to compute an edge score based on a likelihood function modeling noise probability, facilitating robust edge detection and segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If spectral decomposition is performed to extract material information from multi-contrast images, then material-specific images are obtained, but noise is strongly amplified which impedes accurate segmentation
Solution Approach 1:
The patent combines multiple images with different noise characteristics through a unified likelihood function that models noise correlations across images. By merging information from multiple images while accounting for their correlated noise structure, the method extracts material information without the strong noise amplification that occurs in traditional spectral decomposition approaches.
Solution Approach 2:
The patent changes the parameter modeling approach by using a covariance matrix to represent noise correlations between images. This statistical parameter change allows the system to distinguish between signal and noise more effectively, enabling accurate material extraction without noise amplification.
2Difficulty of detecting and measuring
If traditional edge detection methods are used on images with correlated noise, then edges can be detected, but noise correlation and spatial correlation reduce detection accuracy
Solution Approach 1:
The patent changes the detection parameter by using a likelihood ratio that explicitly accounts for noise correlation through a covariance matrix. This parameter transformation converts the problematic correlated noise problem into a solvable statistical discrimination problem, improving edge detection accuracy.
Solution Approach 2:
The patent introduces a covariance matrix as an intermediary that models the noise correlation structure. This intermediary representation allows the system to separate signal from noise by understanding the statistical relationships between different images, thereby improving edge detection precision.
3Loss of information
If multiple images with different contrasts are acquired to encode different physical properties, then more structural information is available, but noise correlation between images complicates processing
Solution Approach 1:
The patent transforms the processing approach by parameterizing the noise model through a covariance matrix. This parameter change simplifies the handling of multiple correlated images by providing a unified statistical framework that automatically accounts for noise correlations, reducing processing complexity despite the increased information content.
Data Source
AI summary
An image processing system comprising: an input port (IN) for receiving two input images acquired of an object. Respective contrast in said images encodes information on different physical properties of the object. The images being converted from a signal detected at a detector (D) of an imaging apparatus (IM). A differentiator of the image processing system forms respective differences from pairs of image points from the respective input images. An edge evaluator (EV) computes, based on said differences, an edge score for at least one of said pairs of image points. The score is based on a measure that represents or is derivable from a conditional noise likelihood function. The likelihood function is based on a probability density that models noise for said signal. Said score is output through an output port (OUT).