遥现中基于显著特征的深度图像滤波算法  

The depth map filter algorithm based on salient features in telepresence

作　　者：冯策[1] 戴树岭[1] 

机构地区：[1]北京航空航天大学自动化与电气工程学院,北京100191

出　　处：《哈尔滨工程大学学报》2014年第11期1364-1368,共5页Journal of Harbin Engineering University

基　　金：国家自然科学基金资助项目(50975010);北京市科委基金资助项目(Z111100074811001)

摘　　要：针对遥现系统中,由于相机精度、材质等原因导致深度信息质量不高,存在缺失以及噪声等问题,提出了基于显著特征的滤波算法。结合频率域的显著图分析算法和基于图的分割算法提取出人类感兴趣的显著区域,针对这部分区域利用联合双边滤波法进行重点处理。实验结果表明,该算法在有效抑制了平坦区域的噪声的同时,增强了显著区域深度图像的细节,并填补了边缘部分的缺失,实现了实时条件下深度图像的有效滤波。In telepresence there are many critical problems, such as low quality depth data, absence of depth information, and noise, which are caused by the precision and material of camera. For solving these problems, a salient features based filter algorithm is proposed. The algorithm first extracts a salient region by the frequency-tuned method and graph-based image segmentation method. Next, the algorithm uses joint bilateral filter to process this region on purpose. Experimental results show that the proposed method can effectively reduce noise, recover the lost region and enhance details at the same time. This allows for achieving filtering depth map effectively in real time.

关 键 词：深度图 联合双边滤波法 显著图 遥现 图像分割 滤波算法 

分 类 号：TN911.73[电子电信—通信与信息系统]