基于背景提取和分区修复的DIBR空洞填补方法

Method Based on Background Extraction and Partition Repair for Hole-filling in DIBR

针对基于深度图的视点绘制中空洞填补时产生的图像失真问题,提出一种还原空洞区域可靠纹理的空洞填补方法。通过从临近的帧画面中提取有效信息还原真实的背景图像,用于新视点中背景区域的空洞填补。并通过区分前景和背景图像区域,对不同区域的空洞采取不同的填充方法,从而获得较为可靠和真实的填补效果。从实验结果来看,该方法获得的PSNR数据优于其他算法,并且与真实视点比较的结构相似度达到99.1%,比传统方法提高10%以上。

For the disocclusion problem occurred in view synthesis using depth map rendering, a new method is proposed to fill the holes with faithful texture in this paper. By extract valid information from adjoining frames,the true background is recovered and it can be used to fill the holes in back- ground areas. By distinguish foreground area and background area, different approuches is used to fill the holes in different areas to get more reliable and ture effect. From the experimental results, PSNR data obtained by this method is superior to other algorithms, and structural similarity reaches 99. 1% compare with the real viewpoint,higher than traditional methods for more than 10%.