The Contracting Curve Density Algorithm: Fitting Parametric Curve Models to Images Using Local Self-Adapting Separation Criteria

摘要: The task of fitting parametric curve models to the boundaries perceptually meaningful image regions is a key problem in computer vision with numerous applications, such as segmentation, pose estimation, object tracking, and 3-D reconstruction. In this article, we propose Contracting Curve Density (CCD) algorithm solution curve-fitting problem. The CCD extends state-of-the-art two important ways. First, it applies novel likelihood function for assessment fit between model data. This can cope highly inhomogeneous regions, because formulated terms local statistics. statistics are learned on fly from vicinity expected curve. They provide therefore locally adapted criteria separating adjacent regions. These replace often used predefined fixed that rely homogeneous or specific edge properties. second contribution use blurred efficient means iteratively optimizing posterior density over possible parameters. enable trade-off conflicting objectives, namely heaving large area convergence achieving high accuracy. We apply several challenging segmentation estimation problems. Our experiments RGB images show achieves level robustness sub-pixel accuracy even presence severe texture, shading, clutter, partial occlusion, strong changes illumination.