| 郭明松,陈芬,汪辉,彭宗举,郁梅,王晓东.基于特征分析的深度视频处理算法[J].光电子激光,2015,26(4):724~732 |
| 基于特征分析的深度视频处理算法 |
| A depth video processing algorithm based on feature analyses |
| 投稿时间：2014-12-05 |
| DOI： |
| 中文关键词: 深度视频 虚拟视点 模糊C均值(FCM)聚类 角点检测 |
| 英文关键词:depth video virtual view fuzzy C-means (FCM) clusteri ng corner detection |
| 基金项目:国家自然科学基金(61171163;61271270;61311140262; U1301257)、 国家科技支撑计划(2012BAH67F01)、“信息与通信工程”浙江省重中之重学科开放基金(xkxl1307)和2014年宁波大学大学生科技创新(SRIP)资助项目 (宁波大学 信息科学与工程学院,浙江 宁波 315211) |
|
| 摘要点击次数: 1083 |
| 全文下载次数: 6 |
| 中文摘要: |
| 在自由视点视频(FVV)系统中,为了提供更真 实的感受,需要通过基于深度的虚拟视点绘制技术获 取足够多的视点,然而由于深度视频获取技术的限制,深度视频的质量很差。为了提高深 度视频的质量, 本文提出了一种基于区域特征的深度视频处理算法。首先,使用模糊C均值(FC M)聚类算 法将深度视频自适应地把深度视频像素分为5类,同时检测深度图的边缘并将边缘块化；其 次,对于非边 缘像素,根据其所在类的特点分别采用改进的三边滤波器及水平方向均值滤波器进行处理； 最后,对彩色 和深度视频分别进行角点检测并块化,并根据边缘像素是否在彩色和深度的角点交集区域内 而采用不同窗 口的中值滤波。实验结果表明,与使用原始深度视频绘制的虚拟视点相比,本文算法能够在 将虚拟视点的客 观质量平均提高0.43dB的同时,显著提高虚拟视点的主观质量；与M artin算法相比,无论是主观质量还是客观质量,本文算法的效果都更加显著。 |
| 英文摘要: |
| In free viewpoint video systems,enough viewpoints are obtained for au dience by depth image based rendering technique,where texture video and corresponding depth video are used.However,the quality of depth video is poor due to the limitation of depth video capture tech nologies.To enhance the quality of depth video,a depth processing algorithm based on feature analyses is proposed in this paper.Firstly,the depth video is adaptively divided into five clusters by fuzzy C-means (FCM) chustering,and the edges of depth video are detected and enlarged at the same time.Secondly,for p ixels in non-edge regions,a novel trilateral filter and a horizontal median filter are respectiv ely adopted according to features of each cluster.Finally,the corners in texture video and depth video are detected and enlarged. For pixel in edge area,median filters with different window sizes are selected according to whether the pixel is in corner area of both texture and depth video or not.Experimental res ults show that the objective quality of the virtual view can be increased by 0.43dB on average compared with that using original depth video while the subjective quality can be improved obviously.No matter in subje ctive quality or objective quality,the algorithm in this paper is much more effi cient than the Martin′s method. |
| 查看全文 下载PDF阅读器 |
| 关闭 |
|
|
|
|
|