Detection of Specularities in Colour Images Using Local Operators

Areas of spectral reflectance, or highlights, can be analyzed for a wide range of information or clues that they give us about a scene. This paper presents a local algorithm for analyzing a moderately unconstrained color image to determine the areas of spectral reflectance. The algorithm is based on the separability of the diffuse and spectral reflection components by differential methods. The location of specular reflectances are marked by finding zero-crossings in concave down regions for twodimensional arrays of intensities representing the color image. These zero-crossings correspond to the centers of the highlight regions. The highlight centers are then expanded to highlight regions by region growing in a direction orthogonal to the local orientation of the highlight. Thus, at the conclusion of the algorithm, the information known about each highlight includes location, size and direction.