Improved gloss depiction using the empirical highlight un-distortion method for 3D image-warping-based stereo rendering

• A highlight un-distortion technique is proposed for 3D image-warped results.
• A variety of highlight distortion patterns under image-based lighting are analyzed.
• A procedure for applying the method to the real-photo scenario is discussed.
• Numerical error and perceptual quality is evaluated to validate the method.

Several methods, such as 3D image warping, have been employed to approximate the stereo effect from one image without full physically based rendering. However, these methods cannot consider the view dependence of highlight depiction, which can undermine the stereo effect. In this paper, a novel method that un-distorts the warped highlights generated by the warping method to reproduce physically plausible highlights is proposed. The nonlinear patterns of highlight distortion are analyzed through a thorough comparison of the warped and physical highlights. The distortion is decomposed into basic and general components, each of which is modeled using a mathematical function. Next, the magnitude and direction of highlight distortion at a point on an object surface are estimated. Spherical approximation of the geometric shape at a point is used to approximately describe the shape of the object. The magnitudes of the basic and general distortion are then obtained using the concept of ‘coupled-sphere’ geometry based on the approximated shape. The directional components of highlight distortion are estimated using the virtual-circle-tracing scheme, which uses a virtual circle that passes a point to obtain the direction for the distortion as the tangent to the circle at that point. Using the magnitudes and directions of highlight distortion, the highlights of the warping method are corrected to produce physically plausible highlights on the surface. The method is fully automatic and computationally efficient. Experimental results indicate that the proposed method greatly improves the visual quality of 3D images.

Figure optionsDownload high-quality image (270 K)Download as PowerPoint slide