We show non-essentiality of using a correlation between an image and a depth map for depth-image-based rendering (DIBR), when we use an accurate depth map. For coding of DIBR, an edge preserving filter, which jointly ...
详细信息
ISBN:
(纸本)9781479947584
We show non-essentiality of using a correlation between an image and a depth map for depth-image-based rendering (DIBR), when we use an accurate depth map. For coding of DIBR, an edge preserving filter, which jointly use the image and the depth map, as a post filter is a suitable approach. The joint filter not only removes coding distortion, but also improves accuracy of the coded depth map itself. Considering the development 3D technology, the accuracy of the input depth map will be improved. If we use an accurate depth map, e.g. the ground truth, the accuracy improvement of the joint filter becomes little. To reveal the fact, we use various codecs (JPEG, JPEG2000, and H.264/AVC) and use two state-of-the-arts filters, which are the post filter set as the non-joint filter, and the weighed mode filter as the joint filter. Experimental results show that the post filters do not require the joint image, and self-sustained types of the non-joint filter have better performance.
In this paper, we will present methods for camera pose estimation for mixed and diminished reality visualization in FTV application. We first present viewpoint Generative Learning (VGL) based on 3D scene model reconst...
详细信息
ISBN:
(纸本)9781479947584
In this paper, we will present methods for camera pose estimation for mixed and diminished reality visualization in FTV application. We first present viewpoint Generative Learning (VGL) based on 3D scene model reconstructed using multiple cameras including RGB-D camera. In VGL, a database of feature descriptors is generated for the 3D scene model to make the pose estimation robust to viewpoint change. Then we introduce an application of VGL to diminished reality. We also present our novel line feature descriptor, LEHF, which is also be applied to a line-based SLAM and improving camera pose estimation.
We have proposed a 3D live video system named LIFLET which stands for Light Field Live with Thousands of Lenslets. It is a computer graphics system based on the optical system of integral photography. It captures a dy...
详细信息
ISBN:
(纸本)0819456373
We have proposed a 3D live video system named LIFLET which stands for Light Field Live with Thousands of Lenslets. It is a computer graphics system based on the optical system of integral photography. It captures a dynamic 3D scene with a camera through an array of lenslets and synthesizes arbitrary views of the scene in real time. Though synthetic views are highly photo-realistic, their quality is limited by the configuration of the optical system and the number of pixels of the camera. This limitation has not been well discussed in our prior works. The contributions of this paper are as follows. First, we introduce a theoretical analysis based on geometrical optics for formulating the upper limit of spatial frequency captured by the system. Second, we propose a system which uses :a combination of an array of lenslets and multiple cameras based on that theoretical analysis. we call it McLiflet since it is a multiple-camera version of LIFLET. The proposed system significantly improves the quality of synthetic views compared with the prior version which uses only one camera. This result confirms our theoretical analysis.
暂无评论