Abstract
In this paper, we present a new processor that, attached to a PC main board, acts as an image associative memory. The main novel contribution is that the operation time required to recognize an object remains constant regardless of the number of stored models. The object silhouette is used as the key image feature allowing recognizing 2D or 3D objects from a single viewpoint. The recognition method is based on a template matching operation in order to simplify the processor architecture and to achieve eight objects per second recognition rate. Digital but also analog techniques have been used in the processor design. Both the architecture and the processing algorithm are presented as well as results with real images.