Improvement On Neighshrink Wavelet Image Denoising Algorithm

Wavelet transform plays an important role in the image processing, especially in image denoising. NeighShrink is an effective noise reduction algorithm which is based on the correlation between the neighboring wavelet coefficients within a subband. It uses the square sum of all the coefficients in the neighborhood to determine how to shrink the center coefficient.In this paper, firstly we summarized different kinds of image denoising methods, with an introduction of the basic theory of wavelet transform, and some classic wavelet denoising algorithms.Secondly, we improved the NeighShrink algorithm by using multiple masks of different shapes and sizes to replace the square neighborhood window. The experimental results show that our method preserves more details of the image, and gets better results in both PSNR measurement and visual quality than NeighShrink algorithm.Then, wavelet coefficients acquired by applying discreet wavelet transform (DWT2) and stationary wavelet transform (SWT2) to nature images are studied systematically. The distribution of wavelet coefficients of different levels and different directions are studied statistically together with the correlation between the center coefficient and the sum of squares of the neighboring coefficients in different neighborhood masks. The results show that both neighborhood method and the use of multi-neighborhood masks have strong statistic bases.Lastly, visual artifacts specific to wavelet image denoising algorithms are studied. A straightforward algorithm to remove these artifacts is proposed. With this method, satisfying visual results can be achieved without compromising the denoising effects.