Depth Image-based Rendering Technology Research Viewpoint

With the development of technology, Free-viewpoint TV has stepped intoour view. Compared with the traditional video technology, FTV can provide usnovel visual experience, which has more interaction and wider visible range.Besides, a fresh immersive can also be provided by such technology, thatbecause two different views’ texture information are sent to our eyesrespectively. Considering that, FTV has more data compared with the traditionalvideo technology which the view is fixed and the visible range is also limited.Hence, it will bring us more pressures on the storage space and transmissionbandwidth. In the face of this, Depth-image Based Rendering (DIBR) isproposed. We can render arbitrary view using several reference views’ textureinformation and their depth. However, the depth’s data is almost20%of texture.The proposed DIBR can reduce3D video’s data largely, which improves thefeasibility of FTV. Based on the learning of DIBR, our researches andinnovations are given as follows:1. Foreground based decision of transition pixels depth is proposed in thispaper, which is used to correct the transition pixels’ depth. Since the depthestimation of transition pixels is unreasonable, it will cause the renderingdistortion in the virtual view. In this method, we generate a threshold using the3D warping formulas and using the threshold to locate transition pixels firstly,and then the depth value of their adjacent foreground is assigned to them.Experimental results show that our method could reduce the distortion ofrendering and improve virtual view’s quality obviously.2. We also propose a region-aware warping for DIBR, which can base thenecessary information on the virtual view. Besides, we only chose the effectiveinformation to do3-D warping and the non-effective and repeated informationwill be abandoned. Since regions are divided reasonably, this idea can realize.With the information of space geometry analyzing, we can conform thereference views’ different regions and then chose the effective information touse. Experimental results show that this method can maintain a high quality of the rendered virtual view. At the same time, its complexity is also reduced.