Kinetic depth images: flexible generation of depth perception

摘要

In this paper we present a systematic approach to create smoothly varying images from a pair of photographs to facilitate enhanced awareness of the depth structure of a given scene. Since our system does not rely on sophisticated display technologies such as stereoscopy or auto-stereoscopy for depth awareness, it (a) is inexpensive and widely accessible, (b) does not suffer from vergence - accommodation fatigue, and (c) works entirely with monocular depth cues. Our approach enhances the depth awareness by optimizing across a number of features such as depth perception, optical flow, saliency, centrality, and disocclusion artifacts. We report the results of user studies that examine the relationship between depth perception, relative velocity, spatial perspective effects, and the positioning of the pivot point and use them when generating kinetic-depth images. We also present a novel depth re-mapping method guided by perceptual relationships based on the results of our user study. We validate our system by presenting a user study that compares the output quality of our proposed method against other existing alternatives on a wide range of images.