Image Rectification for Stereoscopic Visualization

Contact: jinzhou@asu.edu

Abstract-- Stereoscopic visualization can be put to use in many applications, and there exist various methods for this purpose. For example, newly-emerged 3D displays enable convenient 3D viewing without the need for special glasses. In theory, stereoscopic visualization schemes demand a true stereo pair as the input; however, capturing stereo data of natural scenes usually requires special devices such as stereo cameras which are costly and seldom used by common users. To aid the users to obtain the stereo data via common cameras, this paper proposes a systematic approach, which automatically rectifies two images of the same scene captured by common cameras at general positions, so that the results form a stereo pair and thus readily support 3D viewing. Conventional image rectification research primarily focuses on making stereo matching easier but pays little attention to whether the rectified images form a natural stereo pair like those taken by stereo cameras, which is critical to 3D viewing. Unlike conventional methods, the proposed approach rectifies the inputs into a stereo pair, which satisfies the constraints of the stereoscopic visualization platforms. Since physical distortion is minimized in our approach, stereo matching can also be easily facilitated. In addition, the novel derivation of the rectification algorithm has an intuitive physical meaning that is not available from conventional approaches. Further, practical issues related to wide-baseline rectification and operation range of the proposed method are analyzed. To test and validate the proposed approach, both simulated and real data experiments are used and the results are visualized through a 3D display, which is the primary test bed in this study, due to its ease of useDemos: