Advances in Computer Science and its Applications (ACSA) 377 Vol. 2, No. 3, 2013, ISSN 2166-2924 Copyright © World Science Publisher, United States www.worldsciencepublisher.org An Image Denoising Threshold Estimation Method Mantosh Biswas and Hari Om Department of Computer Science & Engineering, Indian School of Mines, Dhanbad, Jharkand-826004, India Email: {mantoshb, hariom4india}@gmail.com Abstract – In this paper, we propose an image denoising threshold method that exploits the subband dependency of the wavelet coefficients to estimate the signal variance using the local neighboring coefficients. The VisuShrink, SureShrink, and BayesShrink denoising methods are important methods for denoising, but these methods remove too many coefficients, leading to poor image quality. The proposed method retains the modified coefficients significantly that result good visual quality. The experimental results show that our method outperforms the VisuShrink, SureShrink, and BayesShrink denoising methods. Keywords – Thresholding; Image Denoising; Peak Signal-to-Noise Ratio (PSNR) 1. Introduction A digital image is more often degraded by noise during its acquisition and/or transmission. It is necessary to remove noise from the image to main its visual quality and it can be done by applying a suitable denoising method. The aim of an image denoising algorithm is to recover the clean image from its noisy version by removing the noise and retaining the maximum possible image information. In the recent years, there has been a fair amount of research on thresholding and threshold selection procedures for image denoising [9-13]. The threshold selection plays an important role in image denoising because the large value of the threshold kills the image data, while the small value of threshold keeps the noisy data [1]. The VisuShrink [1-2], SureShrink [3-4], and BayesShrink [5-6] methods are the most commonly used threshold selection methods. The VisuShrink threshold is a function of noise variance and the number of samples [1-2]. The SureShrink threshold is considered to be optimal in terms of the Stein’s Unbiased Risk Estimator (SURE) [3-4]. This threshold is determined in BayesShrink through modeling the coefficients as Gaussian distribution function [5]. These denoising methods have been improved by our proposed method that follows term- by-term threshold estimation. The rest of the paper is organized as follows. Section 2 gives the overview of the related work. Section 3 describes the proposed denoising method. Experimental results are given in section 4 that is followed by the conclusion in section 5. 2. Related Work There are many threshold selection methods such as VisuShrink, SureShrink, and BayesShrink. The very first time, Donoho and Johnstone gave a mechanism to find the threshold value which is known as VisuShrink [1-2]. The VisuShrink threshold is evaluated by the following expression: T Visu = σ M log 2 (1) where M is the number of pixels in the image and σ is the noise variance that is defined as: σ 2 = [(median| y(i, j) |) /0.6745] 2 (2) here y(i, j)  HH 1 subband coefficients that are obtained by applying the wavelet transform to the image. The VisuShrink has been found to yield an overly smoothed image since the estimate is derived under the constraint with high probability. The SureShrink was proposed by Donoho and Johnstone in which the above problem was overcome using the combination of both the Universal and SureShrink thresholds [3-4]. The SureShrink threshold, T Sure , is defined as: T Sure = min (t J , σ M log 2 ) (3) where t J represents the threshold value at J th decomposition level in wavelet domain. One of the most popular methods namely, BayesShrink was proposed by Chang et al. in which the threshold was derived from Bayesian method [5-6]. This method has better performance than the SureShrink in terms of mean square error (MSE). The BayesShrink threshold for every subband is given as follows: