Computer vision parameter assessment for generic object recognition

Research in computer vision is aimed at making meaningful decisions about scenes of the physical world, based on analyzing images. Segmentation strategies for understanding scenes are one critical step in this process. Scene segmentation is simply the process of attaching symbolic labels to the significant areas in the image of the scene. The particular avenue explored here is based on a novel approach of autonomously directing image acquisition and subsequent segmentation by determining the extent to which surfaces in the scene meet specified functional requirements for generic categories of objects. Results are provided for real data derived from a stereo camera system, the Small Vision System stereo processing software, and the Generic Recognition Using Form and Function object recognition system.