An optimized pixel wise weighting approach for patch-based image denoising
Download as DOCX, PDF0 likes127 views
This document presents a new pixel-wise weighting approach for patch-based image denoising that addresses the limitations of current methods by analyzing the correlation among restored estimates. A bias-variance model is proposed to minimize mean squared error (MSE) through optimization and quadratic programming, enhancing existing algorithms like k-svd and epll. Experimental results indicate an improvement of approximately 0.1 dB in peak signal-to-noise ratio (PSNR) across various noise levels, independent of the image model used.
1 of 1
Download to read offline
More Related Content
What's hot (19)
Ad
Similar to An optimized pixel wise weighting approach for patch-based image denoising (20)
Ad
More from I3E Technologies (20)
Recently uploaded (20)
An optimized pixel wise weighting approach for patch-based image denoising
- 1. AN OPTIMIZED PIXEL-WISE WEIGHTING APPROACH FOR PATCH-BASED IMAGE DENOISING ABSTRACT Most existing patch-based image denoising algorithms filter overlapping image patches and aggregate multipleestimates for the same pixel via weighting. Current weightingapproaches always assume the restored estimates as independentrandom variables, which is inconsistent with the reality. In thisletter, we analyze the correlation among the estimates and proposea bias- variance model to estimate theMean Squared Error (MSE)under various weights. The new model exploits the overlappinginformation of the patches; it then utilizes the optimization to tryto minimize the estimated MSE. Under this model, we proposea new weighting approach based on Quadratic Programming(QP), which can be embedded into various denoising algorithms. Experimental results show that the Peak Signal to Noise Ratio(PSNR) of algorithms like K-SVD and EPLL can be improved byaround 0.1 dB under a range of noise levels. This improvement ispromising, since it is gained independent to which image model isused, especially when the gain from designing new image modelsbecomes less and less.