Image restoration via ising theory and automatic noise estimation

Statistical models, such as the Ising model, have been proven to be very useful in describing solid state systems in physics. Furthermore, these models have been applied to a variety of problems in engineering, chemistry, biology and more, without losing their effectiveness, simplicity and intuitiveness. In this paper we present a clear analogy between the different research fields and a general method for utilizing these models. In a previous work, we introduced a novel Ising-like model and used it in order to restore colored images and videos damaged by various kinds of noise. In this work, we elaborate on important improvements to the restoration algorithm that result in significantly better restorations. Most of the improvements are obtained as a combination of both better physical models and well known image restoration techniques. In particular, the proposed model tests the noisy image automatically and chooses the appropriate model parameters accordingly, on a physical basis, without the need for manual support. Moreover, an automatic analysis of the image histogram is performed, suggesting which pixels are the damaged pixels that need to be restored. In comparison to our previous model, the new model is shown to be not only automatic but also faster. The calculated PSNR and SSIM parameters are better than those achieved previously, as well as by other common filters. Together with the successful results, the disadvantages and limitations of statistical models, such as the Ising model, are discussed as well.