Noise Removal Using Mixtures of Projected Gaussian Scale Mixtures
B. Karthik · 2014
Denoising of natural images is the fundamental and challenging problem of Image processing. As this problem depends upon the type of noise, amount of noise and the type of images it is not truly a practical approach. Wavelet transform method is used for localized in both frequency and spatial domain. The general de-noising wavelet transform method involves three steps 1) Compute the wavelet decomposition of the image 2) Threshold detail coefficients 3) Compute wavelet reconstruction. Dimension reduction methods search for the manifolds in the high-dimensional space on which the data resides. The technique used in this paper for dimension reduction is Principle Component Analysis (PCA). PCA is a well-known technique to map n-dimensional vectors into k-dimensional vectors. a new statistical model for image restoration in which neighbourhoods of wavelet subbands are modeled by a discrete mixture of linear projected Gaussian Scale Mixtures (MPGSM). In each projection, a lower dimensional approximation of the local neighbourhood is obtained, thereby modeling the strongest correlations in that neighbourhood. The algorithm used is Expectation Maximisation (EM) algorithm.