完全基于边缘信息的快速图像分割算法

A Fast Image Segmentation Algorithm Fully Based on Edge Information

为快速准确地进行图像分割和对象提取,达到图像实时处理的要求,提出一种完全基于边缘信息、具有良好并行计算特性的图像分割和对象提取算法.首先对经典Canny算子边缘检测进行改进、引入了断点检测和递归可控的低阈值边缘追踪算法,得到图像的原始二值边缘图;然后采用基于断点的边缘连接算法得到待提取对象的完整闭合边缘;最后提出四向扫描区域填充算法从边缘图中完整地提取出有意义的对象.实验结果表明,文中算法不仅可以准确地处理图像中的各类复杂边缘、获得良好的图像分割效果,同时具有优秀的并行计算特性,可以以并行方式、基于GPU实现完全实时的图像处理.

In order to get quick and accurate image segmentation and object extraction for real-time applications, a new algorithm which is fully based on edge information is proposed in this paper. In this new algorithm, Canny edge detection, with improvements in edge detection and edge tracing for parallel processing, is used to generate an edge map of the original image. Then a new edge-connection method is introduced to connect breakpoints and therefore to obtain closed contours of target objects. Finally, a new region filling method called four-direction-scanning is raised to fulfill object extraction. The whole algorithm is optimized for parallel calculation. Experiments show that the proposed algorithm performs an accurate execution of object extraction and image segmentation for any kind of complicated images. Moreover, throughout its full exploitation of GPU＇s parallel calculation resources, the proposed algorithm has a dramatic reduction of its runtime which enables its use in today＇s real-time applications.