Team Imagine of the laboratory LIRIS is aimed to analyze the image contents through segmentation of the region(s) of interest, feature extraction by the bias of dense descriptors so as, for example, to identify or characterize the content(s) of the objet(s) in the image. The proposed Ph.D thesis topic is exactly in the same context, since it concerns analyzing photos of coin in order to provide a computer-aided authentication and evaluation.
More precisely, the company G.E.N.I., which will sell a protection case for coins of collection, would like to hold appraisal sessions of coins in large scale. It expects to grade 1000 to 1500 coins per day during one week’s appraisal session. That is the reason why the company seeks the competences of LIRIS to automate this process. From photos taken in the well-controlled conditions (lighting and camera angle controlled), we are asked to extract the coin region from the image, to recognize its form, to detect its material and its facial value, but above all, to provide an automatic assistance to its authentication and grading. More concretely, the authentication is carried out by comparing theoretical features related to the size, with those of the authenticated ones. The grading is estimated according to wear, to any contingent scratches and hits that the coin could have, but also to the visibility of details on the targeted elements, to its patina or its luster.
This application raises several scientific challenges which will be studied during the research work composed of four steps. The first step consists of segmentation and feature extraction in order to estimate the size and the form (round, number of sides or teeth, atypical) of the coin. The second step is addressed to extract “expected” interior elements in the coin. In the third step, the texture of the coin, in general sense, will be studied to determine its material and to quantize its luster. The forth step will focus on analyzing the “non-expected” elements inside the coin. More precisely, it refers to locating contingent hits or scratches inside the coin, as well as contingent traces of wear.