Semi-automatic Extraction of Parametric Characteristic Features from Images

Abstract

In order to extract exact characteristic features for industrial design, this paper proposed a novel semi-automatic approach for abstracting high quality parametric contours from images. Using prior knowledge and parameterized feature template for given objects, our method combines global matching/ morphing with hierarchical part-based approximation. The method starts from construction of parametric templates associated target object. Through contour detection preprocessing and template retrieval, we obtain the exact morphable feature template. Combining edge-oriented contour detection, corner detection and local prior template, the exact endpoints and curve segments are obtained. Finally, the segmented parametric feature lines are automatically fitted using cubic B¡äezier curves. Taking car body design as numerical examples, we establish the corresponding database. Numerical results show that our method is effective to abstract the design features with high precision for car styling.