IMPACT OF WAVELET DECOMPOSITION SCALE AND SIMPLEX THRESHOLDING ON IMAGE DENOISING

Abstract

Image denoising is a pivotal problem in many image processing applications (segmentation, denoising, feature extraction, compression and so on). An effective wavelet-based image denoising mainly depends on the selection of an appropriate wavelet basis, decomposition to the appropriate scale, estimation of an optimal threshold, and function over the wavelet coefficients. Therefore, in this paper, at first, we report in detail the potential trends, such as the choice of wavelet basis and wavelet decomposition to the proper scale. Orthogonal, biorthogonal and reverse biorthogonal classes of various discrete wavelet filters are treated with different structured benchmark test images at various additive Gaussian noise levels. Also, we analysed the impact of PSNR in dB between each decomposition scale, which discloses the choice of resolution levels. Finally, we proposed a simple exponential (simplex) thresholding strategy to preserve the image features well. Our method is tested with a Bayesian framework and compared with state-of-the-art thresholding methodologies involving traditional and hybrid wavelet coefficient thresholding. Experiments show that our system is feasible in terms of simplicity and outperforms the state-of-the-art filtering techniques.