Abstract:
Image de-noising based on wavelet can be extended to video de-noising by using it on each frame separately. By manipulating inter-frame correlations de-noising performance can be enhanced by using appropriate temporal filtering. Fixed temporal filtering may not give appropriate results due to their inability to deal with the variations of inter-frame correlations. Many adaptive temporal filtering methods for de-nosing in spatial domain exist, but they do not directly outspread in wavelet-based de-noising.
In scalar Hidden Markov Tree, prior state probabilities are plugged into algorithm to estimate conditional probability density function. Hidden Markov tree modeling vector extension in wavelet domain is proposed, that exploits the frame reliance of wavelet coefficients. This will estimate unknown parameters by using expectation maximization algorithm.
Experimental results reveal that the vector estimation of wavelet coefficients gives better de-noising performance as compared to existing techniques, in terms of quantitative and qualitative analysis
