Contextual hidden Markov model-based image denoising in sharp frequency localized Contourlet transform domain

A contextual hidden Markov model (CHMM) for image denoising application was presented in sharp frequency localized Contourlet transform (SFLCT) domain. Firstly, cycle spinning technology was employed on the noisy image, and then decomposed by the SFLCT into sub-images, solving one drawback of the original Contourlet transform that its basis images were not localized in the frequency domain and compensating for the lack of translation invariance property of SFLCT, suppressing the pseudo-Gibbs phenomena around singularities of images. Secondly, a new context design scheme was proposed, CHMM was established aiming at high frequency subband coefficients and applied to image denoising. Finally, the denoised image was reconstructed by inverse SFLCT and inverse cycle spinning operation. Valid transform mechanism and a comprehensive statistical correlative model that was constructed by integrating context information with HMM were utilized, which could fully express persistence across scales, directional selectivity within scales and energy concentration in the space neighborhood of Contourlet coefficients, consequently, the proposed method was more effective and suitable for image denoising. Some experiments are conducted to verify the method is more potential and certainly superior to wavelet transform method and the original Contourlet transform method, both in subjective evaluation and numerical guidelines.