Image Denoising By Mixing Wavelet Transformation With Sparse 3D Collaborative Filtering

Image denoising refers to the use of various filter models, including the traditional filtering, wavelets, partial differential equations and other methods known to remove the noise from the noisy image. Image denoising belongs to the image preprocessing stage from the view of image analysis's flow,and belongs to the image restoration from the view of the digital image processing technology. Its existence has very important significance.As it has compact support and its decomposition has stratification, wavelet has been widely used in research and practice.It mainly use the fact that image signal and noise signal present different rule through wavelet transformation at different resolutions, by adjusting the wavelet coefficients to achieve the purpose of image denoising.The algorithm of image denoising by sparse 3D transform-domain collaborative filtering (BM3D) has made a great achievement in image denoising. However, because it processes 2D images by 3D groups, the computation amount may be very large, especially when finding the similar blocks for reference blocks. An improved algorithm is proposed in this paper. Not only does it keep the good performance of the BM3D algorithm, but also it can reduce the computation time significantly. We realize it using the three successive steps: wavelet decomposition of an image, process in the wavelet transformation domain by using the BM3D method, and wavelet reconstruction. An efficient implantation of this algorithm is presented in full detail. Also given is the comparison of this improved algorithm with the BM3D approach. The experimental results demonstrae that this improved method achieves excellent denoising performance in terms of even higher signal-to-noise ratio, subjective visual quality and much less computation amount.