Deep learning for patch based denoising has done best performance but if an image is comprised of various similar patterns then the performance of these CNNs gets degraded. Persistent homology is a mathematical model based on topological analysis of data method. This paper proposes a novel method to de-noise an image using Persistent homology residual learning for block match three dimension algorithm using a deep residual learning algorithm is used with feature space. The learning incorporated here is mainly for the performance to be improved by the input and tag manifolds which in turn is simpler in terms of topologically for mapping to a feature space. From the experimental conducted and the obtained results its demonstrate the effectiveness of the persistent residual learning in image de-noising for block match three dimension algorithm and it is observed that the proposed algorithm outperforms in terms of elimination of Gaussian noise in images based on performance metric and visual quality.

No abstract

BM3D is a recent state of art patch based denoising algorithm .It works on the fact that an image has a locally sparse representation in transform domain.It is composed of two stages i)hard thresholding and ii)weiner filtering This paper provides a mechanism that incorporates an improved version of BM3D which combines the digital image characteristic with added noise pollution levels, and adaptively selects block-matching threshold in grouping stage for an extended BM3D to four dimension so as to denoise volumetric data corrupted by Gaussian and rician noise. Experimental results demonstrate it outperforms not only in terms of objective criteria of PSNR, but also in improving the visual quality.