Restoration of ancient colored documents foreground/background separation

In this paper, we propose a new approach for foreground/background separation to restore degraded ancient documents suffering from recto/verso transparency. This approach is used to process single-sided documents and proceeds in three steps: The first step aims to separate between the background and foreground of the image by a contrast enhancement using gamma correction and histogram stretching. The second step consists in separating the final foreground and the remaining noised pixels. This step is based on the pixel-level characterization of the initial foreground followed by a Principal Component Analysis. Then, the results are ameliorated using a Fuzzy C-Means classification with two classes followed by a maximum likelihood algorithm to refine the results and to correct the misclassified pixels. Finally, the third step allows the reconstruction of the final restored image using the obtained foreground and background. The new approach presents the advantage of the accuracy of its results.