Removal Algorithm Of Gaussian Noise In Video Sequence

During the acquisition and transmission process of video sequence, it often introducenoise because of some uncontrollable factors, which affects people’s subjective visualquality and brings difficulties to the subsequent further processing of the video sequence.Remove noise in video sequence has become a hot topic in the field of image processingresearch.In the single image denoising algorithms, Block Matching and3D Filtering algorithm(BM3D) incorporate the non-local block matching ideas into the empirical wiener filtering.It achieves good denoising effect. Based on the correlation between the video sequencedenoising and the single image denoising, we extend the BM3D algorithm to the videosequence denoising. For each frame in the video is divided into certain overlapping imageblocks with the same size, taking each block as a reference successively and finding alimited number of blocks that are most similar to the reference block from previously andsubsequent frames. To form a3D group based on the similarity in descending order. Thendeal with the group by hard-thresholding filtering which based on fourier transform, afterthat, obtain the estimates of the blocks in the group and determine the group’s estimateweight according to the compactness of the blocks in the group that before filtering. Afterdealt with all the reference blocks, getting the basic estimate of the entire video using theweighted average of all overlapping block-estimates. Divided a set of blocks from thebasic estimate video using the above method and form a “clean group” for each block. Foreach “clean video”, extracting the blocks whose locations are the same as the blocks in“clean group” from noisy video to form the “noisy group”. Designing empirical wienerfiltering based on the “clean group” and filtering the “noisy group”. Finally, for blocks ofeach group in the filtering result weighted for the final estimates of the video.Four groups of video having different characteristics from standard video sequencelibrary has been adopted, we make simulation experiments in the videos added with zeromean white gaussian noise with different intensity. The experiments show that thedenoising effect is good.