Abstract Discriminating edge types, based on their local surface reflectance properties, is useful for a number of applications such as object recognition, stereo vision and structure from motion, where similar edge types (e.g. material transitions) from two distinct images are used for image matching while discounting other "accidental" edge types (e.g. shadows and highlight transitions). Because intensity-based edge detectors cannot distinguish between various transition types (that is whether the transition is due to material changes, shadows, abrupt surface orientation changes or highlights), in this paper, we aim at using color information to classify the physical nature of the edge.
Therefore, the effect of varying imaging circumstances is analyzed. From this analysis we present the color models c1c2c3 and l1l2l3. It is shown that l1l2l3 varies with a change in material only, c1c2c3 with a change in material and highlights, and RGB vary with a change in material, highlights and geometry of an object. From these color models we derive gradient information which is used to classify edges in a color image to be one of the following types: (1) a shadow or geometry edge, (2) a highlight edge, (3) a material edge.
Experiments conducted with the edge classiffication technique on different color images show that the proposed method successfully discriminates the three different edge types.