基于区域中心的交互式图像前景提取方法

图像分割是从图像中提取有意义的区域,是图像处理和计算机视觉中的关键技术。而自动分割方法不能很好地处理前景复杂的图像,对此提出一种基于区域中心的交互式图像前景提取算法。针对图像前景的复杂度,很难用单一的相似区域描述前景,文中采用多个区域中心来刻画目标区域。为提升图像分割的稳定性,给出基于超像素颜色、空间位置和纹理信息的相似性度量方法;为确保图像分割区域的连通性和准确性,定义了基于超像素的测地距离计算方法。使用基于测地距离的超像素局部密度,来分析图像的若干区域中心;基于用户交互的方式来分析前景的区域中心,得到图像前景。经过大量彩色图像的仿真表明,在分割过程中利用少量的用户交互信息,可有效提升图像分割的稳定性和准确性。

An Interactive Image Foreground Extraction Method Based on Region Center

Image segmentation is the extraction of meaningful regions from images and is a key technology in image processing and computer vision.The automatic segmentation method can’t handle the complex foreground image well,so we propose an interactive image foreground extraction algorithm based on region center.For the complexity of image foreground,it is difficult to describe the foreground with a single similar area.We use multiple regional centers to describe the target area.In order to improve the stability of image segmentation,a similarity measure method based on superpixel color,spatial position and texture information is given.To ensure the connectivity and accuracy of image segmentation region,a geodesic distance calculation method based on superpixel is defined.The local density of the superpixel based on the geodesic distance is used to analyze several regional centers of the image;the regional center of the foreground is analyzed based on the user interaction to obtain the image foreground.The simulation of a large number of color images shows that a small amount of user interaction information is used in the segmentation process,which can effectively improve the stability and accuracy of image segmentation.