Gaussian Adaptive Filtering of Low Resolution Video Using Anisotropic Tensor

In this paper an optimization procedure is proposed for finding the optimal parameters of the Anisotropic Gaussian filter for low resolution video, containing Additive White Gaussian Noise, using anisotropic tensor. The tensor is found from the video content by estimating the brightness gradient along horizontal and vertical direction, according to the algorithm of Kimmel et al. High quality of the filtered frames is achieved, corresponding to Peak signal to Noise Ratio of 30.07 dB and Structural Similarity Index Measure of 0.8641, when the input video contains noise with variance, equal to 0.001. Comparative analysis with the filter of Middendorf et al., employing also the structure tensor, along with the non-adaptive Average and Gaussian filters, reveal better structural preservation of the video, processed by the optimized version of the proposed filter.