Image enhancement using wavelet-domain mixture models

We propose a non-linear mapping function for digital image enhancement in the wavelet domain, which amplifies midrange coefficients more than small and large coefficients. We derive this function based on a statistical model of the wavelet coefficients. This three-component mixture model describes the coefficients in each subband as a mixture of small, medium, and large coefficients to which different amplification factors are assigned. The model parameters are estimated from each subband using the EM algorithm. The algorithm has a small number of user-specified parameters while can obtain good enhancement results.