Wavelet Domain Hmt Model-based Image Denoising

Image denoising is very important in the field of the digital image processing. Traditional way to denoise can make the edge of the image faint. Translation invariant is lacked in the general way to denoise using wavelet transform, so the Gibbs in the denoised image happens often. Besides, because the general way to denoise using wavelet transform doesn't use the dependence of the wavelet coefficients, the result of it isn't very good. In order to solve this problem, the image denoising technology based on wavelet HMT is researched in this text.A new method suppressing Gaussian noise based wavelet HMT is proposed in the text .we shift the image horizontally, vertically and diagonally first, then use the Hidden Markov Tree model (HMT) to characterize the wavelet coefficients of the shifted image. The 2-state Gaussian mixture model is used to approximate the image wavelet coefficients margin distribution and the matrix of state transform can capture the dependencies of the wavelet coefficients across the scales, then in order to get the parameters of HMT ,EM method is used to train the HMT. While the wavelet HMT is used in image denoising, the wavelet coefficients of noised image need matching the HMT to get the wavelet model of image signal and Bayesian is used to estimate the wavelet coefficients of real image, we use the estimated wavelet coefficients reconstruct the denoised image .Then we shift the corresponding denoised images contradictorily; choose the average of the all denoised images as the last denoised image. The simulation results show that the denoising method can wipe off the white noise better and increase the PSNR while containing the detail of the image edge and weakening the Gibbs effectively.Besides, a method that image denoising of multilevel median filter based on threshold decomposition combined with HMT in wavelet domain is proposed in this text in order to suppress mixed noises. The simulation tests show that the method preserves the details and edges of image better while suppressing the mixed noise well.