Chapter 8. Ground from Figure Discrimination 8.5 Conclusions Chapter 8. Ground from Figure Discrimination Chapter 8. Ground from Figure Discrimination

completion elds: A neural model of illusory contour shape and salience. eferences 123 OPR78] R. Ohlander, K. Price, and D. R. Reddy. Picture segmentation using a recursive region splitting method. Sau92] E. Saund. Labeling of curvilinear structure across scales by token grouping. In CVPR, pages 257{263, 1992. SB93] S. Sarkar and K. L. Boyer. Perceptual organization in computer vision: a review and proposal for a classifactory structure. SU88] A. Sha'ashua and S. Ullman. Structural saliency: The detection of globally salient structures using locally connected network. In ICCV-88, pages 321{327, 1988. SU90] A. Sha'ashua and S. Ullman. Grouping contours by iterated pairing net-cuts within twice the optimal. eferences 121 HvdH93] F. Heitger and R. von der Heydt. A computational model of neural contour processing: gure-ground segregation and illusory contours. Three-dimensional grouping and information fusion for site modeling from aerial images. Got91] C. Gotsman. A cluster detection algorithm based on percolation theory. Quantitative analysis of the viewpoint cosistency constraint in model-based vision. algorithm for nd-ing best matches in logarithmic expected time. eferences 117 their increasing complexity, as this has never been proved. However, in this thesis we provide a supporting example in which a general model for the representation of perceived organization information is used for all these tasks | creating a saliency map, gure ground discrimination and grouping. In this model, the complexity of the proposed corresponding algorithms increases in the same hierarchical order. Finally, we believe that various perceptual grouping processes should be combined in order to make a complete system. Ground from gure would be the rst ltering stage, followed by edge completion and perceptual grouping. All these stages can use the same data structure and the same grouping cues as their source of information, but may also use diierent cues at diierent stages. This is not a new idea, and we believe the proposed framework to be a convenient tool for combining all these tasks. References AB95] T. Alter and R. Basri. Extracting salient curves from images: An analysis of the saliency network. every grouping task. We have shown that often, it is possible to obtain more reliable cues (with predicted reliability) with relatively little computational eeort. This procedure incorporates multi-feature cues into the grouping process. One way to look at grouping cues is as quantiiers of high order statistical information (higher order than 1st order statistics). A straightforward example is the co-linearity cue described above, which examines …