Vector Quantizer for Image Estimation in Wavelet Domain

摘要

An image denoising algorithm operating in the redundant wavelet transform domain is proposed. A Vector Quantizer is trained over the wavelet coefficients of the noisy image. The algorithm is comprised of a spatially adaptive Bayesian estimation procedure used in conjunction with the VQ at lower scales of the wavelet transform domain. This algorithm requires a priori knowledge of the shape of the pdf (for Bayesian estimation ) of the wavelet coefficients. However, the training of the VQ suggested herein is based on the co-efficients of the noisy image, facilitating the application of VQ independent of the statistical characteristics of the image under consideration. Further, the smoothness constraint for physical images is being assumed.