3.1 Methodology
The main assumption in designing the verification approach is that the segmentation of the words into characters carried out by the HRS is reliable. This assumption is based on previous studies that have shown that the segmentation is reliable in most of the cases [9], [31], [32]. For instance, a visual inspection carried out on 10,006 word images from the training data set of the SRTP1 database together with a comparison with the alignment (segments-characters) provided by the HRS has identified segmentation problems in less than 20 percent of the cases (2,001 out of 10,006 words).