Classi cation of Color Edges in Video intoShadow-Geometry , Highlight , or Material

We aim at using color information to classify the physical nature of a color edge. To achieve physics-based edge classiication, we propose a taxonomy of color invariant edges. Further, we present a framework for automatic color edge detection and noise-adaptive thresholding derived from sensor noise analysis and propagation. Then, a parameter-free color edge classiier is introduced labeling color transitions into the following edge types: (1) shadow, geometry or shading edges, (2) highlight edges, (3) material edges. The proposed method is empirically veriied for video showing complex real world scenes.