Image Filter Based on Block Matching, Discrete Cosine Transform and Principal Component Analysis

An algorithm for filtering the images contaminated by additive white Gaussian noise is proposed. The algorithm uses the groups of Hadamard transformed patches of discrete cosine coefficients to reject noisy components according to Wiener filtering approach. The groups of patches are found by the proposed block similarity search algorithm of reduced complexity performed on block patches in transform domain. When the noise variance is small, the proposed filter uses an additional stage based on principal component analysis; otherwise the experimental Wiener filtering is performed. The obtained filtering results are compared to the state of the art filters in terms of peak signal-to-noise ratio and structure similarity index. It is shown that the proposed algorithm is competitive in terms of signal to noise ratio and almost in all cases is superior to the state of the art filters in terms of structure similarity.