CLASSIFICATION OF TEXTURED SURFACES BASED ON REFLECTION DATA (IMAGE-PROCESSING, COMPUTER VISION)

Recognition of surfaces is very important in computer vision. Color and texture of a surface are two key characteristics that influence the surface recognition process. Light reflected from a surface contains information about these two characteristics. The color of a surface determines how a surface reflects light of a specific wavelength, and the surface texture determines the amount of diffuse and specular reflection taking place from the surface.^ A statistical approach has been developed for classification of surfaces based on the spatial intensity distribution of light reflected from the surfaces. In this approach, a surface is modelled as a collection of mirror-like micro-facets oriented randomly with respect to each other. The random orientations of the micro-facets give rise to spatial intensity distribution of the reflected light. The reflected light field is characterized by a probability density function, which is used for deriving first and second order features. The first order features are the mean and variance of the reflected intensities, while the second order features are based on the spatial correlation between the reflected intensities. A correlation matrix is formulated based on the co-occurrence of two given intensities separated by a given angular distance. Two classification schemes based on maximum-likelihood and nearest-neighbor decision rules are implemented on the feature sets.^ Experimental results for the classification schemes are presented for a variety of sample surfaces. These include paper, cloth, felt, sandpaper, cork, crumpled and smooth aluminum foils, etc. The success rate of 80-100% has been achieved for the two classification schemes.^ In addition to these statistical features, other properties of textures such as gloss, contrast, roughness, etc., have also been measured from the reflected intensities. Parameters such as micro-facet slope function, micro-facet orientation function, etc., have been measured and used for predicting the statistics of the perceived texture. Techniques have been developed to reduce the computational burden associated with this reflection data based surface classification scheme in order to make it suitable for on-line operation.