ABSTRACT
The proposed algorithm is an improvisation of the multichannel image denoising where the fine details in the frame sequences are handled by the multiscale non-local spatial filter and the coarser details are handled by curvelet transform with hard thresholding. The algorithm consists of three stages: fragmentation of the video sequences to respective 2D frames, noisy pixel identification in the 2D frames and denoising the pixels to obtain original pixels. The process begins with the processing of video sequences for conversion to 2D frames and removal of obsolete frames from the sequence. In the next stage a multichannel color processing with non-local spatial filter scheme deals with the flat regions and the fine details in the frame sequences and coarser details are handled with the curvelet transform to separate the actual object tracked and the noise subspace. Finally, the actual denoised frame sequence is recovered to obtain the video for estimating the motion of multiple objects in frame by preserving the edges and details. The proposed approach is experimented on some benchmark data sets, and its performance is evaluated using two performance evaluation measures: SSNR (Subtractive Signal to noise ratio) and MPSSI (Mean Preservation Speckle Suppression Index). The proposed approach gives better results that state-of-the-art techniques.
