A new approach for the removal of mixed noise based on wavelet transform
Yufeng Li, HongXia Ni, Wei Pang, ZhiZhang Hao · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2006
This paper proposed a new approach for the removal of mixed noise. There are many different ways in image denoising. Donoho et al have proposed a method for image de-noising by thresholding, and indeed, the application of their method to image denoising has been extremely successful. But the method of Donoho is based on the assumption that the type of noise is only additive Gaussian noise, which is not successful for impulse noise. Mallat has also presented a method for signal denoising by discriminating the noise and the signal singularities through an analysis of their wavelet transform modulus maxima (WTMM). Nevertheless, the tracing of WTMM is not just tedious procedure computationally; ambiguity is often resulted in determining the correspondence of a modulus maximum to a singularity. In the light, we combine the merits of the two techniques to form a new approach for the removal of mixed noise. At first, we used wavelet singularity detection (WSD) technique to analyze singularities of signal and noise. According to the characteristic that wavelet transform modulus maxima of impulse noise rapidly decreases as the scale increases in wavelet domain, it can be accurately located with multiscale space by going through dyadic orthogonal wavelet transform and removed. Furthermore the Gaussian noise is also removed through a level-dependent thresholding algorithm. The experimental results demonstrate that the proposed method can effectively detect impulse noise and remove almost all of the noise while preserve image details very well.