VIDEO DENOISING BASED ON EXPLOITING INTRASCALE AND INTERSCALE DEPENDENCY IN WAVELET DOMAIN

In this paper, we propose a new video denoising algorithm which uses an efficient wavelet based spatio-temporal filter. The filter first applies 2D discrete wavelet transform (DWT) in horizontal and vertical directions on an input noisy video frame and then applies 1-D discrete cosine transform (DCT) in the temporal direction in order to reduce the redundancies which exist among the wavelet coefficients in the temporal direction. We observe that the subband coefficients with large magnitudes occur in clusters in locations corresponding to the edge locations even after applying the above spatiotemporal filter. In this paper, we propose to use two different low complexity wavelet shrinkage based methods to denoise the noisy wavelet coefficients in different subbands. The first method exploits the intra-scale dependencies between the coefficients and thresholds the wavelet coefficients based on the measure of sum of squares of all wavelet coefficients within a square neighborhood window. The second method exploits the inter-scale dependencies between the coefficients at different scales in an individual slice of coefficients. After filtering the individual slices of coefficients, the denoised video frames in time domain are obtained after inverse transforms. We propose to exploit the temporal redundancies between the successive frames again in the time domain using low complexity selective recursive temporal filtering (SRTF). In the proposed video denoising scheme, since the temporal redundancy is exploited both in the time and wavelet domain, the denoising capability of the scheme is hence increased. The video denoising performance using the two proposed approaches outperform many existing well known video denoising techniques including one recent well known method which uses the similar transformation, both in terms of PSNR and visual quality. We also show that, simple soft thresholding using Donoho’s threshold when used with this wavelet based spatio-temporal filter even outperforms many well known non linear based video denoising techniques