Quality Assessment of 3D-Synthesized Images

Abstract

AbstractIn recent years, the technologies related to 3D-synthesized images, such as free viewpoint video and 3D television, have attracted considerable attention due to their extensive applications in fields like distance education, entertainment, and remote monitoring. The new viewpoints generated by the depth image-based rendering (DIBR) technique are synthesized in reference-free image environments (in “blind” environments), so there is an urgent need for no-reference quality assessment (QA) methods of 3D-synthesized images. However, most traditional no-reference QA models fail to effectively measure the geometric distortions caused by DIBR. For the 3D-synthesized image QA, this chapter first introduces the no-reference QA methods on the basis of autoregressive modeling and multi-scale natural scene statistical analysis to measure geometric distortion. Second, it describes the no-reference QA methods on the basis of pixel-based changes in transform domains to capture color and depth distortion. Third, it presents no-reference QA methods based on the structural variations caused by geometric, sharpness, and color distortions to evaluate the blurred, discontinuous, and stretched 3D-synthesized image quality. Finally, the above-mentioned QA methods are validated on relevant databases, and the necessity of proposing efficient 3D-synthesized image QA methods is pointed out.