Reflection Decomposition in Single Images Using an Optimum Thresholding-based Method

Abstract

Traditional methods of separating reflection components have been developed based on multiple images. There are only few methods which are able to use a single image. However, their applicability is limited due to offline setting of its arbitrary parameter. In this study, we propose an effective method to separate specular components using a single image which based on an optimum thresholding-based technique. This method employs modified specular-free image and selects an optimum value for the offset parameter. In contrast to prior method, the proposed method processes all the steps automatically and produces better performance. Experimental results for inhomogeneous objects demonstrate the promising applicability for real-time implementation. However, this method is unsuitable for objects with strong specular reflection. An extension is suggested to include the specular lobe reflectance into Shafer dichromatic model.