We introduce a new shape descriptor, the shape context, for measuring shape similarity and recovering point correspondences. The shape context describes the coarse arrangement of the shape with respect to a point inside or on the boundary of the shape. We use the shape context as a vector-valued attribute in a bipartite graph matching framework. Our proposed method makes use of a relatively small number of sample points selected from the set of detected edges; no special landmarks or keypoints are necessary. Tolerance and/or invariance to common image transformations are available within our framework. Using examples involving both silhouettes and edge images, we demonstrate how the solution to the graph matching problem provides us with correspondences and a dissimilarity score that can be used for object recognition and similarity-based retrieval.