融合彩色信息和深度信息的GrabCut图像分割  

GRABCUT IMAGE SEGMENTATION COMBINING COLOR AND DEPTH INFORMATION

作　　者：凌滨[1] 郭也 赵永辉[1] 李超[1] Ling Bin;Guo Ye;Zhao Yonghui;Li Chao(School of Mechanical and Electrical Engineering,Northeast Forestry University,Harbin 150036,Heilongjiang,China)

机构地区：[1]东北林业大学机电工程学院,黑龙江哈尔滨150036

出　　处：《计算机应用与软件》2020年第8期188-193,共6页Computer Applications and Software

基　　金：国家自然科学基金项目(31700643);中央高校基本科研业务费专项资金项目(2572018BF15)。

摘　　要：图像分割是计算机视觉的重要组成部分,但大多数图像分割工作主要集中在二维图像的处理上,故结合深度信息和GrabCut算法提出一种新的分割方法。为了准确地分割出目标物体,利用物体深度信息的同时提出一种基于背景和前景先验的GrabCut图像分割方法。融入深度信息,选择深度特征结合流形排序算法来构造图模型;为了进一步突出目标对象,抑制背景区域,分别利用背景先验和前景先验,生成相应的显著图,将二者融合并进行优化得到最终待处理图像;以深度信息指导GrabCut算法进行精分割,得到分割结果。实验结果表明,该方法能够较为准确地分割出目标对象。Image segmentation is an important part of computer vision,but most image segmentation work mainly focuses on the processing of two-dimensional images.To this end,we propose a new segmentation method combining depth information and GrabCut algorithm.In order to segment the target object accurately,we propose a GrabCut image segmentation method based on background and foreground prior while using the depth information of the object.Taking the depth information into account,we chose the depth feature and the manifold sorting algorithm to construct the graph model;to further highlight the target object and suppress the background area,we used the background prior and foreground prior respectively to generate the corresponding saliency map,and the final image to be processed was obtained by fusion and optimization;the depth information was used to guide the GrabCut algorithm to perform fine segmentation,and the segmentation results were obtained.The experimental results show that our method can accurately segment the target object.

关 键 词：图像分割 深度信息 GrabCut算法 背景先验 前景先验 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]