Part Segment at ion for 0 b ject Recognit ion

Thas paper addresses the problem of segmentang objects anto paris iising stereo amages. There are three components an the pnrt segmentation process: surface segmentaiaon, regaon groupang, and ziolumetrac models recovery/segmentataon. The surface segmentataon process segmenis the amage anto a set of regaons such that each regzon represents a surface smooth zn depth. The regaon groupang process merges the segmented regaons anto rnennangful part. Fanally the process of volumetrac model recovery/segmeniaizon phase recovers the part model and segments that part aiito smaller parts zf necessary. Sance we use mul-ta-shape models to drave the pari segmentataon process, we can capture more geometrac propertaes of the object and the applacaiaon domnan of our approach as broader ihan that of the prevaous approaches.

[1]  Alex Pentland,et al.  Perceptual Organization and the Representation of Natural Form , 1986, Artif. Intell..

[2]  L.-H. Chen,et al.  Surface segmentation from stereo , 1992, Proceedings 1992 IEEE Computer Society Conference on Computer Vision and Pattern Recognition.

[3]  A. Pentland Recognition by Parts , 1987 .