Image Denoising Based on Wavelet Transform and BM3D Algorithm

Abstract

Image denoising as a key method is innovating continuously. Since the Block matching and 3D (BM3D) algorithm is superior to other methods in suppressing Gaussian noise, it has become the current state-of-the-art of denoising. Nevertheless, the image detail information will be partially lost during eliminating image additive noise, there is still room for improvement. Aiming at the existing problems of BM3D algorithm, an improved BM3D algorithm is designed by a combination of wavelet transform and BM3D algorithm. The method applies the principle that wavelet denoising preserves fine edge information to compensate the missing edge details caused by BM3D algorithm. The wavelet threshold denoising runs in parallel with the BM3D algorithm. Firstly, the wavelet threshold denoising method is used to obtain the preprocessed image. Meanwhile, the BM3D algorithm is applied to the corrupted image, including the basic estimate and the final estimate operation, to get another denoising preprocessed image. Finally, the final result comes out from pixel-level averaging of the two preprocessed images. Experimental analysis on three images of Lena, Barbara and Cameraman illustrate that denoising, using the proposed algorithm, can provide largest value of peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM). The qualitative and quantitative research results illuminate that the improved algorithm is effective and robust.