Image denoising using hidden Markov models

In this paper a new method is proposed for image denoising based on HMM (hidden Markov modeling). In this manner the noisy imageis modeled as a hidden Markov process, we use the local statistics of the images in defining the HMM parameters, and the image restoration is achieved by computing the most likely state at each pixel. Among the features of the proposed method is the adaptive window size for different regions of the image (smooth and nonsmooth regions). The adaptive window size allows us to obtain a better estimate of the local variance of the noise, therefore, the additive noise is removed more in the smooth regions while the edges are preserved in nonsmooth ones. Another feature of this method has to do with the proportionality of the execution time and the noise power; the less the noise power, the faster is the execution of the proposed algorithm. The performance of this algorithm is evaluated through subjective and objective criteria and it is shown that the restored images by HMM have higher contrast and clearness which is attributed to nearly optimal usage of the statistical properties of the image by HMM.