A New SURE Approach to Image Denoising : Interscale Orthonormal Wavelet Thresholding Florian Luisier, Thierry Blu, Senior Member, IEEE, and Michael Unser, Fellow, IEEE IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 16, NO. 3, MARCH 2007 1 Outline • • • • • Introduction SURE Approach to Image Denoising PSNR Comparisons and Visual Quality Computation Time CONCLUSION 2 Introduction • A nonredundant transform may match the performance of redundant ones. • Do not make any explicit hypotheses on the clean image. • Near-optimal performance—both regarding quality and CPU requirement 3 SURE Approach to Image Denoising(1/5) Our goal is to find a function that minimizes By Stein’s Lemma and leads to 4 SURE Approach to Image Denoising(2/5) The sensitivity of the soft-thresholding function with respect to the value of T is high. 5 SURE Approach to Image Denoising(3/5) Build a linearly parameterized denoising function of the form This linear system is solved for a by 6 SURE Approach to Image Denoising(4/5) 7 SURE Approach to Image Denoising(5/5) The number of terms K and the parameter T can be fixed independently of the image. 8 Interscale Predictor 9 Interscale Predictor 10 PSNR Comparisons and Visual Quality 11 PSNR Comparisons and Visual Quality 12 PSNR Comparisons and Visual Quality • 2 important criteria of judging visual quality are widely used: – The visibility of processing artifacts can be reduced by taking into account intrascale dependencies – The conservation of image edges can be reduced by a careful consideration of interscale dependencies in the denoising function 13 Computation Time 14 CONCLUSION • Demonstrate the efficiency of our SURE-based approach (best output PSNRs for most of the images). • The visual quality of our denoised images is moreover characterized by fewer artifacts than the other methods. 15
© Copyright 2026 Paperzz