Speckle Noise Reduction in SAR and Ultrasonic Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques fail to effectively reduce speckle noise and segment coherent images, such as those from synthetic aperture radar and ultrasonic systems, due to assumptions about multiplicative noise and reliance on user input, which degrades image quality and diagnostic value.
Innovation Solution
The method employs a Markov Random Field (MRF) algorithm to establish a coherence factor, noise threshold, and pixel threshold, performing uniformity tests and intensity updates within a neighborhood system to reduce speckle noise and segment images based on conditional probability functions, suitable for real-time processing in SAR and ultrasonic systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If conventional speckle noise reduction techniques (Lee filter, Kuan filter, Frost filter) are used, then some noise reduction is achieved, but they assume multiplicative noise which prevents substantial eradication of speckle noise
Solution Approach 1:
The patent transforms the image from intensity domain to log-complex domain, changing the mathematical representation parameters. This allows the noise model to be changed from multiplicative to additive, enabling more effective noise reduction while preserving image quality through subsequent filtering operations in the transformed domain
Solution Approach 2:
The patent replaces conventional filtering mechanisms (Lee, Kuan, Frost filters) with a wavelet-based processing system. This substitution uses wavelet transformation and thresholding mechanisms instead of traditional statistical filtering, achieving superior speckle noise reduction while maintaining diagnostic image quality
2Extent of automation
If automated image segmentation is attempted using conventional techniques (edge detection, region growing, thresholding), then segmentation can be performed, but these techniques require user input and are adversely affected by speckle noise
Solution Approach 1:
The patent applies speckle noise reduction processing before segmentation operations. By preprocessing the image to remove noise while preserving edges and features, subsequent automated segmentation algorithms can operate more reliably without user intervention and achieve higher accuracy
Solution Approach 2:
The patent uses wavelet transformation to decompose the image into different frequency subbands, enabling multi-resolution analysis. This segmentation in the frequency domain allows automated processing to operate on specific frequency components, improving both automation capability and segmentation accuracy
Data Source
AI summary
The present invention includes methods for the reduction of speckle noise in an image and methods for segmenting an image. Each of the methods disclosed herein includes steps for analyzing the uniformity of a pixel within a plurality of pixels forming a portion of the image and, based on the uniformity of the intensity of the plurality of pixels, adjusting and/or replacing the pixel in order to produce a speckle-noise reduced image, a segmented image, or a segmented and speckle-noise reduced image. The methods of the present invention can employ for example conditional probability density functions, nonlinear estimator functions, convex energy functions and simulated annealing algorithms in the performance of their respective steps.