计算机工程与应用 ›› 2016, Vol. 52 ›› Issue (2): 186-190.

• 图形图像处理 • 上一篇    下一篇

基于SLIC的改进GrabCut彩色图像快速分割

胡志立,郭  敏   

  1. 陕西师范大学 计算机科学学院,西安 710062
  • 出版日期:2016-01-15 发布日期:2016-01-28

Fast segmentation in color image based on SLIC and GrabCut

HU Zhili, GUO Min   

  1. School of Computer Science, Shaanxi Normal University, Xi’an 710062, China
  • Online:2016-01-15 Published:2016-01-28

摘要: GrabCut算法用户交互量少且分割精度高,但它迭代使用GraphCuts的求解模式使得在处理高分辨率图像时,耗时巨大。提出了一种快速GrabCut算法,在高斯混合模型参数估计过程中,通过SLIC算法构建精简的GraphCuts模型以实现加速。通过SLIC算法将原始图像快速地预分割成具有确定边界且区域内相似度高的超像素图,并以此构建精简的网络图。以块内的RGB均值描述超像素特征进行高斯混合模型参数估计。为了提高分割精度,使用得到的GMM参数对原始图像进行分割。实验结果证明了该算法在时效和精度上都有很好的性能。

关键词: 简单线性迭代聚类(SLIC), 图割, 高斯混合模型

Abstract: GrabCut requires less user interaction and segmentation with high accuracy, but it is an iterative solver mode using GraphCuts which costs enormous time when processing high resolution images. This paper presents a fast GrabCut algorithm. To achieve acceleration, a simplified GraphCuts?model is constructed by SLIC algorithm in parameter estimation of Gauss mixture model. The original image?is?pre-segmented into super-pixel image with determining boundaries and internal high similarity through SLIC algorithm, a simplified network graph is constructed by this. The Gaussian Mixture Model (GMM) parameters can be estimated with the mean RGB values within blocks instead of all pixel values. To improve segmentation accuracy greatly, the GMM parameters obtained by the last step in parameter estimation of?Gauss mixture model are used to segment the original image. The experimental results demonstrate this algorithm with segmentation accuracy and computation efficiency.

Key words: Simple Linear Iterative Clustering(SLIC), graph cuts, Gaussian mixture model