A Study On Stereo-plus-Depth Imaging And Visual Comfort Enhancement For Stereoscopic 3D Images

With the rapid development of stereoscopic 3D(S3D) display technology, more and more people begin to pay more attention to stereoscopic images and video. It give us a more life-like and visually immersive viewing experience. However, it becomes the one of challenges should be overcome that the viewing safety of S3 D contents. In viewing process, visual discomfort and visual fatigue are induces by many factors such as excessive disparity which beyond the adjusted range, the mismatches in the left and right view image, the accommodation-vergence conflicts, etc. In this thesis, firstly, a configuration of the stereoscopic image capturing system is proposed for our research team, which can obtain 3D images and corresponding disparity map accurately and effectively. And then, based on a new concept model about visual attention and visual discomfort, we introduce an effective way to assess and visualize visual discomfort. Finally, we propose a visual comfort enhancement using saliency-adaptive nonlinear disparity mapping method. The main contents of the thesis can be summarized as follows:1. The increasing number of people pays more attention to stereoscopic contents. Therefore, how to achieve stereoscopic image with high quality and strong sense of stereo is become a new tendency. Based on the studies of current stereoscopic image capturing system, we describe a proposed stereo-plus-depth image capturing system. It is composed of one trinocular camera and one TOF depth camera, which can capture stereoscopic image in three types and the corresponding depth image.2. In the process of viewing stereo contents, we discover that the visual attention and visual discomfort is determined by different factors. Based on the HVS, the image saliency is related to the visual attention, it determines which region has been focused by human eyes. Meanwhile, the disparity saliency is related to visual comfort, the larger excessive disparity give uncomfortable experience to viewers. According to the new concept model, we obtain the salient discomfort region in an image by threshold segmentation, and extract two disparity features to assess the visual discomfort. Then, based on the predicted visual comfort score which denote the degree of visual comfort of an image, and then we visualize the salient discomfort region. Experimental results demonstrate that the proposed method can significantly achieve higher accuracy in visual discomfort evaluation, and have high correlation with the subjective assessment.3. The thesis proposes a saliency-adaptive nonlinear disparity mapping for visual comfort enhancement in stereoscopic 3D images. The new combination model of the saliency information and disparity information is proposed, which can introduce saliency information efficiently without substantial change in the original disparity. Based on the analysis of human visual system(HVS), the viewers usually are attracted by the salient regions. Thus, we compress salient discomfort regions more, non-salient discomfort regions less, and both salient and non-salient comfort regions little. Experimental results shows that the proposed method can enhance visual comfort by adjusting the excessive disparity while retaining the overall viewing image quality.