摘要
针对光场相机图片的前后景自动分割算法研究,采用了恢复光场相机图片深度图,结合RGB图和深度图两者的优点,然后使用图割算法,实现了光场图片前后景自动分割。首先计算光场图片的深度图且对光场图片的RGB图进行超像素处理,然后依据深度图确定可信的前后景区域和模糊区域,接着对可信的前后景区域进行K-means聚类,最后使用图割算法自动分割光场图片前后景。由于使用了深度图,文中不需要任何用户交互的先验知识,即能有效地自动分割光场图片前后景。大量的实验显示了我们自动分割算法的有效性。
In this paper,we propose a binary image segmentation method on light field images.The method takes advantage of both RGB and depth information generated by post-processing images from a light field camera.We first pre-process the image pair including RGB image and depth image,then transform image pixels into super pixels,next identify the confident foreground and background together with the undetermined area and obtain the confident areas by K-means,finally run Graph Cut to gain the best segmentation for the unknown area.By using depth information,we no longer require user input as a prior information.A large number of experiments show the effectiveness of our automatic segmentation algorithm.
作者
徐小杰
柳畅
XU Xiao-jie;LIU Chang(Shanghai Institute of Microsystem and Information Technology,Chinese Academy of Sciences,Shanghai 200050,China;ShanghaiTech University,Shanghai 201210,China;University of Chinese Academy of Sciences,Beijing 100049,China)
出处
《电子设计工程》
2019年第4期136-139,145,共5页
Electronic Design Engineering
关键词
光场
前后景分割
图割
超像素
light field
foreground segmentation
graph cut
super pixels
