We develop a score to compare a full or partial shoe outsole impression (Q) from an unknown source to an impression (K) from a known reference shoe. The data consist of the coordinates of all edges detected on each of the images. Degree of correspondence between Q and K will be quantified for several local regions of Q. We rely on graph theory and a maximum clique (MC) approach to quantify the degree of correspondence. An advantage of the MC method is its invariance to rotation and translation. For each local region of Q, we calculate a rotation angle that results in best alignment with a local region of K. We also calculate the number of overlapping points after alignment. Information from multiple local regions will be combined to construct a score using a suitable machine learning algorithm on ground truth known reference data collection. Initial results on a small sample of footwear impressions are promising. The next step is to explore the behavior of the score when it is computed for images that are known to have the same source and for images known to arise from different sources.