Resorting to restoration techniques for heritage documents becomes an increasingly urgent need. In fact, these valuable resources for human being are subject to several types of degradations limiting their use. A proposed solution to this problem is the application of restoration techniques on the digital copy of the originally degraded document. This would improve human readability and allow further application of image processing techniques. Hence identifying a typology of different types of image degradation is of primary concern for use in restoration techniques. Our main contributions in this paper are twofold. In the first instance, we propose a typology for different types of degradation of old document images. Our proposed typology is lead by the type of image processing undertaken in the course of virtual restoration. The second contribution is to develop a restoration method treating specific document degradation: "ink bleed-through". The proposed method is a non-supervised segmentation method. It is based on a recursive segmentation approach applied to the principal component analysis space. As an illustration, a scheme of "ink bleed-through" removal of the provided document images by the archive of "Chatillon-Chalaronne" is presented. Experiments conducted on these real ancient document images illustrate the effectiveness of our proposed method.