Additive noise reduction in natural images using second-generation wavelet transform hidden Markov models

Noise reduction or denoising is required for visual improvement or as a preprocessing step for subsequent processing tasks, such as image compression and analysis. Therefore, denoising has become a highly desirable and essential process in multimedia applications. The aim of all denoising processes, especially in natural images, is to uncover the true image from the observed noisy image, ideally removing the additive white Gaussian noise (AWGN) while producing a sharp, useful image without losing finer details. Generally, most of the noise obtained during acquisition and transmission of the natural images is assumed to be AWGN. In this study, we propose a new adaptive denoising framework based on second‐generation wavelet domain using hidden Markov models (SGWD‐HMMs) with some new local structure, utilizing the fact that the images are nonstationary with singularities and some smooth areas that can be considered as stationary. The dependencies among wavelet coefficients can be efficiently captured by HMMs since the dependence between two wavelet coefficients dies down quickly as their distance becomes big. Quite remarkably, experimental results verify the effectiveness of SGWD‐HMMs in benchmark images when compared with other state‐of‐the‐art denoising algorithms. It gives competitive results in the subjective and objective assessments, but it is computationally more expensive. © 2016 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.