Entropy-based image noise variance estimation

In the de-noising and segmentation algorithm used to deal with the noise image, it is necessary to know the distribution model and the statistical parameters of noise. A novel noise estimation algorithm was thus proposed. First, the combined value of the input noise image variance and local entropy of each image block was calculated. Then all the comprehensive values were arranged in a descending order, and de-noising was calculated using the corresponding standards deviations in that order. Finally, final noise estimates were selected using the image quality evaluation algorithm. The proposed algorithm does not need pre-processing such as complex filtering, wavelet transform, etc., and can obtain the variance of noise by directly processing a series of input image data. It is simple and easy to implement, has high computational efficiency, and enable BM3D and similar de-noising algorithm to denoise adaptively.