270 likes | 529 Views
Switching Bilateral Filter With a Texture/Noise Detector for Universal Noise Removal. Chih-Hsing Lin, Jia-Shiuan Tsai, and Ching-Te Chiu Transactions on: Image Processing, IEEE Journals 2010. Outline. Introduction Sorted Quadrant Median Vector for Noise Detection Noise models
E N D
Switching Bilateral Filter With a Texture/Noise Detector for Universal Noise Removal Chih-Hsing Lin, Jia-Shiuan Tsai, and Ching-Te Chiu Transactions on: Image Processing, IEEE Journals 2010
Outline • Introduction • Sorted Quadrant Median Vector for Noise Detection • Noise models • Definition of Sorted Quadrant Median Vector (SQMV) • Features of SQMV • Edge/Texture identification with the clusters of SQMV • Reference median • Switching Bilateral Filter • Switching scheme • Noise detector design • Switching bilateral filter • Experimental Results • Conclusions
Introduction • Gaussian noise: a zero-mean Gaussian distribution. • Effective filter: linear filters (ex: averaging) • Side effect: blurring • Impulse noise: replacing a portion of an image pixels with noise values. • Effective filter: nonlinear filters (ex: median) • In this paper, we propose a universal noise removal filter based upon the “detect and replace” methodology.
-Noise models • The Impulse noise corrupted pixel ui,j: • Salt-and-pepper: ni,jonly takes values of Lmin or Lmax. • Uniform impulse: ni,jtakes random values from the interval [Lmin , Lmax]with a uniform distribution. • The Gaussian noise corrupted pixelui,j: • In this paper, mixed impulse and Gaussiannoiseis considered, and the Gaussian noise is independent of impulse noise.
Sorted Quadrant Median Vector for Noise Detection The processing window size is too small. • Motivation of the Noise Detection Scheme: • Existing two-state noise detectors fail in several conditions[9][17]. • The central pixel of (a)(b)identified as noise-freepixel. • The medians of(c) stillsimilar. [9] T. Chen and H. R. Wu, “Adaptive impulse detection using center-weighted median filters,” IEEE Signal Process. Lett., vol. 8, no. 1, pp. 1–3, Jun. 2001. [17] P. E. Ng and K. K. Ma, “A switching median filter with boundary discriminative noise detection for extremely corrupted images,” IEEE Trans. Image Process., vol. 15, no. 6, pp. 1506–1516, Jun. 2006.
-Definition of Sorted Quadrant Median Vector (SQMV) • To overcome the problems, we propose a sorted quadrant median vector (SQMV): • For a (2N+1) *(2N+1) window we divide the window into four (N+1)*(N+1) subwindows. • In the case N = 2:
-Definition of Sorted Quadrant Median Vector (SQMV) • The set of points can be expressed as: • For (2N+1) *(2N+1) window: • For (N+1) *(N+1) subwindows: • Where the SQMV is defined as: • SQM1, SQM2, SQM3 and SQM4 are the medians m1, m2, m3, and m4 sortedin an ascending order.
-Edge/Texture identification with the clusters of SQMV • The differencebetween two boundary values: ρ lies in the interval [25–40]
-Edge/Texture identification with the clusters of SQMV • Experimental result:
-Reference median • In “without edge” or “weak edge” cases, the reference median (SQMR) for xij is the average of SQM2 and SQM3 (major cluster). • In “edge or texture” case, decide which cluster the current pixel xij falls into by dav:
-Reference median • The pixel selection of x1~x4: • Thereference median (SQMR)in each case: Even if complextexture , the filtering result would be less artificial. “without edge” or “weak edge” “edge or texture”
Switching Bilateral Filter • Bilateral Filter: • xi,j: the current pixel ̶yi,j: the filtered pixel • xi+s,j+t: he pixels in (2N+1)*(2N+1) window
-Switching scheme • In the switching scheme, we the noise detector searches for noisy pixels and tries to distinguish them from uncorruptedones. • The filtered image is defined as follows: • S1 and S2: the binary control signals generated by the noise detector.
-Noise detector design • The noise detection : • The threshold: • For salt-and-pepperimpulse noise: [Tk1 Tk2] =[3015] • For uniform impulse and Gaussiannoise: [Tk1 Tk2] =[255]
-Switching bilateral filter • Propose a new universal noise removalalgorithm: the switching bilateral filter (SBF) • Parameter selection: • For “edge” σS= 3, otherwise σS = 1. • σR= [30,50] will work well, we choose σR= 40.
Conclusions • Propose SQMV for edge/texture detection, noise detection and switching bilateral filter. • The noise detector shows a good performance in identifying noise even in mixed noise models. • In most of the noise model cases, proposed filter outperforms both in PSNR and visually.