Denoising Natural Images Using Fast Multiscale Directional Filter Banks
Epiphany Jebamalar Leavline, Shunmugam Sutha · International journal of tomography and simulation · 2015
This paper presents a novel denoising algorithm for additive white Gaussian noise removal in natural images. This algorithm uses a subband selection technique for selecting the suitable subband for the estimation of the threshold value for noise removal. Fixing the optimum threshold is a key factor that determines the performance of any denoising algorithm. Fast multiscale directional filter bank (FMDFB) is a perfect reconstruction framework suitable for a several image processing applications because of its directionality and less computational complexity. Unlike wavelet subbands, each subband in FMDFB carries significant information about the content of the image. Hence, a simple wavelet-based subband adaptive shrinkage method tends to oversmooth the FMDFB coefficients. To overcome this issue, the proposed algorithm selects the suitable subband for the estimation of the threshold value based on the statistical properties of the subbands. The experimental results show that the proposed FMDFB based denoising algorithm outperforms the wavelet and contourlet based denoising for Gaussian noise removal.