基于GrabCut的改进分割算法  

An improved segmentation algorithm based on GrabCut

作　　者：王茜 何小海[1] 吴晓红[1] 吴小强[1] 滕奇志[1] Wang Qian;He Xiaohai;Wu Xiaohong;Wu Xiaoqiang;Teng Qizhi(Institute of Image Information,School of Electronics and Information Engineering,Sichuan University,Chengdu 610065,China)

机构地区：[1]四川大学电子信息学院图像信息研究所,四川成都610065

出　　处：《信息技术与网络安全》2021年第10期43-47,52,共6页Information Technology and Network Security

基　　金：国家自然科学基金(62071315)。

摘　　要：针对GrabCut算法对于特征不明显、纹理复杂的图像分割效果不理想,且需要用户交互的问题,提出一种基于GrabCut的改进分割算法。首先,运用图像增强,对特征不明显的图像进行改善,提高图像质量;然后,利用YOLOv4网络对图像进行目标检测,获取前景目标所在矩形框位置,从而减少用户操作;其次,在高斯混合模型(GMM)中加入图像像素的位置信息和局部二值模式算子(LBP)提取的像素纹理特征信息,优化高斯混合模型参数,改进GrabCut算法,实现图像优化分割;最后,将分割图像掩膜与原始图像结合,得到原始图像。实验结果表明,对特征不明显、纹理信息复杂的图像,该算法分割效果更优。To slove the problem that GrabCut does not have satisfactory segmentation effect for images with obscure features and complex textures and it needs user interaction,an improved segmentation algorithm based on GrabCut was proposed.Firstly,image enhancement was used,to improve the image with less detailed features.Secondly,YOLOv4 network was trained and the image was put in YOLOv4 to get the rectangular position of the foreground target.Thirdly,Gaussian Mixing Model(GMM)was incorporated location information of image pixels and texture feature information extracted by LBP operator,to optimize GMM model parameters and improve GrabCut algorithm.Finally,the original segmented image was obtained by combining the segmented image mask with the original image.The experimental results show that the proposed method performs better on images with less detailed features and complex texture information.

关 键 词：GRABCUT K-MEANS 图像增强 图像分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]