Discrete TV Minimization Denoising for Non-Uniform Noise
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image denoising techniques using total variation (TV) minimization often overcompensate in low noise regions and fail to effectively reduce noise in pixel clusters, leading to excessive smoothing or loss of spatial resolution, especially in images with non-uniform noise distributions common in clinical applications.
Innovation Solution
The implementation of a Total Variation Iterative Reconstruction (TV-IR) method that employs discrete TV minimization with a one-direction condition, which adjusts pixel values based on gradients to reduce noise in high noise regions without overcompensating in low noise areas, and uses ordered subset simultaneous algebraic reconstruction technique (OSSART) to process projection data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If total variation minimization is applied to reduce noise, then noise is reduced uniformly, but edges are smoothed out and spatial resolution is sacrificed
Solution Approach 1:
The patent applies different smoothing strengths to different regions of the image based on local noise characteristics. High noise regions receive stronger smoothing while low noise regions and edges receive minimal or no smoothing. This is achieved by calculating a noise metric for each region and using it to weight the TV minimization operation, thereby preserving edges while reducing noise where needed.
Solution Approach 2:
The patent dynamically adjusts the TV minimization parameter (lambda) based on local noise levels. The parameter is increased in high noise regions to strengthen denoising and decreased or set to zero in low noise regions and edge areas. This adaptive parameter adjustment allows the algorithm to reduce noise globally while preserving local edge sharpness and spatial resolution.
2Object-affected harmful factors
If total variation minimization is applied to reduce noise, then noise is reduced, but overcompensation occurs in low noise regions causing excessive smoothing
Solution Approach 1:
The patent evaluates local noise metrics and applies TV minimization selectively. Regions identified as having low noise or containing edges are excluded from strong smoothing operations, while only high noise regions undergo denoising. This local differentiation prevents overcompensation and preserves original pixel values in areas where they should remain intact.
Solution Approach 2:
The patent applies partial TV minimization rather than full minimization across the entire image. By limiting the smoothing operation to only those regions where noise exceeds a threshold, the algorithm avoids excessive action in low noise regions. This partial application ensures that original pixel values are preserved where noise is already low, preventing information loss.
3Object-affected harmful factors
If total variation minimization is applied to reduce noise, then noise is reduced, but odd pixel clusters remain as black/white dots due to cluster effect
Solution Approach 1:
The patent adjusts the TV minimization parameter dynamically based on local conditions. In regions containing odd pixel clusters (identified by high noise metrics), the parameter is increased to strengthen the denoising effect. This allows the algorithm to overcome the cluster effect and eliminate isolated black/white dots while maintaining stability in uniform regions.
Solution Approach 2:
The patent incorporates feedback mechanisms that monitor noise reduction progress and adapt the minimization strength accordingly. When odd pixel clusters are detected through noise metric analysis, the algorithm increases smoothing strength in those specific regions. This feedback-driven adaptation ensures reliable denoising of difficult-to-process cluster regions while avoiding unnecessary smoothing elsewhere.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method of denoising an image, comprising the steps of a) initializing (S100) a direction index associated with each of image pixels, the direction index indicating an increase, a decrease and a mixed state in gradient direction of total variation at a given pixel, b) determining (S110) a discrete gradient of the total variation for each of the image pixels, the discrete gradient indicating the increase, the decrease or the mixed state, c) updating (S130) the direction index of each of the image pixels with the discrete gradient if the direction index is the mixed state, d) changing (S150) a pixel value of each of the image pixels by a predetermined value to reduce the total variation if the direction index and the discrete gradient are equal in the gradient direction for the one of the image pixels, and e) repeating (S160, S170) the steps b) through d) until a certain predetermined condition is reached.