Regularized variational dynamic stochastic resonance method for enhancement of dark and low-contrast image

Dynamic stochastic resonance (DSR) is a distinctive technique for enhancement of dark and low-contrast image. Noise is necessary for DSR based image enhancement and the level of noise will be enlarged simultaneously with brightness, which reduces the perceptual quality of the enhanced image greatly and also increases the difficulty of subsequent denoising because removing high level of noise often leads to serious loss of image details. In this paper, instead of removing noise after the enhancement process is complete, we propose to suppress noise gradually and simultaneously in the process of enhancement. We rewrite the traditional partial differential equation (PDE) based DSR model in variational framework firstly, and then propose a novel total variation regularized (TV) DSR method for image enhancement. The existence and uniqueness of solution of the TV regularized DSR model is proved theoretically. Moreover, we generalize the TV regularized DSR model in variational framework and in PDE framework, respectively, and therefore we can incorporate more existing denoising methods into our approach. Numerical comparisons demonstrate that the proposed technique gives significant performance in terms of contrast and brightness enhancement as well as noise suppression, and therefore can obtain enhanced image with good perceptual quality.