In this paper we propose a shape recognition approach applied to a dataset composed of 512 shoeprints where shapes are strongly occluded. We provide a local adaptation of the HRT (Histogram Radon Transform) descriptor. A shoeprint is decomposed into its connect components and describes locally by the local HRT. Then, following this description, we find the best local matching between the connected components and the similarity between two images is defined as mean of local similarity measures.