RECONSTRUCTION OF SATELLITE IMAGE CONTAMINATED BY CLOUDS

The presence of clouds within a satellite image limits their analysis of data for application in different fields. Therefore, the elimination of clouds in satellite images is a subject widely studied to solve this problem. One of the techniques used for the reconstruction of images is the inpainting, which consists in restoring a damaged area in visually plausible form using information outside the damaged domain. The proposed method is to use the Expectation-Maximization algorithm for the classical case of Gaussian mixtures, to find the classes (labels) within an image. For the process of searching for the most similar patch, it is proposed to use the sum square error (SSE) both for the brightness of pixels as well as for the classes (modes) found in the image. Combining the classes found in each neighborhood (patch) with the color information allows us to obtain a better description of the patches. Some experiments were conducted to compare the results with other methods presented in the literature.