Skip to Main Content
In object recognition problems, we may face the situation in which two similar objects with very different angles are mistaken to be two different objects. Here we present a novel approach to solve this kind of problem by emphasizing the shape recognition. In our framework, the measurement of similarity is preceded by (1) solving correspondences between points on the two shapes, (2) using the correspondences to estimate an aligning transform, (3) after transformation, we use the PMK to estimate the similarity of the two shapes. In order to solve the correspondence problem, we attach a descriptor, the shape context, to each point. The shape context at a reference point captures the distribution of the remaining points related to it, thus offering a globally discriminative characterization. Corresponding points on two similar shapes will have similar shape contexts, which enables us to solve correspondences as an optimal assignment problem. Given the point correspondences, we estimate the transformation which best aligns the two shapes; then we use kernel-based classification method--pyramid match kernel to estimate the similarity between two shapes.