Recovering Reflections Of Three-dimensional Object Surfaces

For dielectric inhomogeneous objects, the perceived reflections are the linear combinations of diffuse and specular reflection components. The specular reflection component, which is also called highlight, is depended on the position, shape and the surface roughness of the object and the orientation of the light source. It will affect the accuracy of many algorithms in image processing. The separation of diffuse and specular reflection components, or equivalently specularity removal, plays an important role in the fields of image analysis, pattern recognition, and scene synthesis. This issue has been studied in recent years. Several methods have been presented based on image segmentation or local interaction of neighboring pixels.This thesis first introduces several related techniques based on single image, multi-views, and color line for the separation of reflection components, and then proposes a simple and effective specularity-removal method on the level of each individual pixel. The chromaticity of diffuse reflection is approximately estimated by employing the concept of modified specular-free image, and the specular component is adjusted according to the criterion of smooth color transition along the boundary of diffuse and specular regions. This method does not require image segmentation and even local interactions between neighboring pixels. Experimental results indicate that the proposed method is promising when compared with other state-of-the-art techniques, in both separation accuracy and running speed.The reflection separation algorithm proposed has broad usage. It can be the base of other methods for diffuse image analysis and processing, and the separated specular component can be used in estimation of surface roughness and lighting direction, or for image-based scene synthesis.