| Summary: | 碩士 === 國立臺灣科技大學 === 資訊管理系 === 104 === As the techniques in stereo vision can provide user diversity experience, it becomes more popular now, and can also be used in various fields. One of these techniques called multi-view autostereoscopic display. It provides user immersive experience. In addition, it does not need any wearable device to achieve three-dimensional effect. This technique provides an easier way for amateur users. Currently there are many creations of wiggle stereoscopy, and the creation can be shared easily. However, if the image content is not processed properly, the stereoscopic effect cannot be shown very well.
This paper proposes a flat automultiscopic display by simulating the views where a person should see with different viewing angles for a fixed position. Motion parallax provides the depth cue from images[1]. Given an image sequence we make some rectification to remove artifacts like motion blur and stereo 3D ghosting. Also we take a human visual system into account to enhance automultiscopic displays’ effect. Our system adopts the idea of Changjae Oh et al[5]. Utilize “Probabilistic Correspondence Matching using Random Walk with Restart”. Once we get the disparity map, we repair the invalid region which contain error matching in the disparity map, and do the rendering check, so that we can reduce image blur and stereo 3D ghosting. At last, we using saliency map to decide fixation of human vision and make image alignment.
|
|---|
