Research Of Denosing Algorithm Especial For Video Encoder

The problem of video denoising has been deeply discussed constantly in the past several decades.In some cases such as realtime video surveillance,realtime streaming media service, mobile visual phone and so on, we must find comparatively simple denoising method with fine performance in order to satisfy both the quality and instantaneity, which can be coupled with the encoder tightly.This paper analyzes and generalizes the existing algorithms on video denoising theoretically,and focuses on the transform domain wiener filter algorithm and transform domain time filter algorithm especially,which both are coupled tightly into the coder. Both of the two algorithms are inefficient under low profile T264 encoder environment, and this can be proved by testing and comparing their denoising performances with no noise reduction algorithm.Under the basic profile circumstance of T264 encoder,the transform domain weiner filter algorithm can't denoising properly.To solve this problem,this paper suggests that when calculating the coefficient covariance array of DCT dormain,only the direct coefficient variance equals 1,and all the other variances are equals to the first alternating current coefficient's variance. Experimental results indicate that under the basic profile circumstance of T264 encoder and with known noise intensity,to video sequences with lower speed and more flat areas,the improved transform wiener filter is optimal in compact ratio and objective quality when compared with other algorithms.And it still has comparatively stable denoising performance when the noise intensity is increased.Under the basic profile circumstance of T264 encoder, the filter factor generated by the transform domain time algorithm is often close to the thresholds that hinder denoising properly. To solve this problem,this paper presents a new hierarchical non-linear method for determining the filter factor which is suitable for H.264 encoder.It selects relevant empirical filter factor for a 4×4 block according to the sum of absolute difference and motion vector firstly,and then implements denoising by multiplying the filter factor with each DCT coefficient. Experimental results indicate that under the basic profile circumstance of T264 encoder,and if the noise intensity is known,the improved transform dormain time filter can gain correspondingly finer denosing effect by setting proper branches and relevant thresholds.Since the improved method doesn't need to estimate the noise variance, it's complexity is lower than the original algorithm.