Cognitively Motivated Shape Similarity

L.J. Latecki and R. Lakaemper (USA)


shape similarity, visual parts, shape representation, object recognition


In this paper we present a novel approach to shape based object recognition. In accord with human visual percep tion, it is sufficient to recognize the shape of part of an object in order to identify the object if the shape of the part is distinctive. The proposed approach follows this principle and works for objects extracted from real images. Assum ing a visible part is distinctive, we can find objects with this part even if only this part of a given object is visible and the visible part is significantly distorted. The main contri bution of this paper can be made clear using an analogy to text-based retrieval. Existing shape similarity measures force the users to submit a description of the whole ob ject shape (e.g., the whole object contour) as query, which corresponds to submitting the whole sentence as the query, making it very unlikely to find a good match. In contrast, the proposed partial similarity allows the users to submit key parts of shape (parts of object contours) as query in analogy to keywords in the text based search.

Important Links:

Go Back