摘要： 基于图割理论的GrabCut算法由于使用所有像素来迭代估计高斯混合模型(GMM)参数，算法效率较低。针对该问题，提出一种基于图割的JPEG图像快速分割算法。以GrabCut算法为基础，对JPEG图像中DC系数构成的低频图像进行迭代分割，估计GMM参数以减少训练样本的数目。实验结果表明，该算法能在保证分割精度的前提下缩短高分辨率JPEG图像的分割时间。

关键词: 图割, GrabCut算法, 高斯混合模型, JPEG标准, 离散余弦变换, 直流系数

Abstract: GrabCut algorithm based on graph cuts is less efficient because it uses the whole pixels to estimate Gaussian Mixture Model(GMM) parameters by iteration. Aiming at this problem, this paper proposes a fast JPEG image segmentation algorithm based on graph cuts. On the basis of GrabCut algorithm, the Direct Current(DC) coefficients of JPEG image that constitute the low-frequency image are used to iteratively estimate the GMM parameters, which greatly reduce the number of training samples. Experimental results show that this algorithm can shorten the segmentation time of high-resolution JPEG image while preserving the segmentation accuracy.

Key words: graph cut, GrabCut algorithm, Gaussian Mixture Model(GMM), JPEG standard, Discrete Cosine Transform(DCT), Direct Current (DC) coefficient

中图分类号: 

• TP391.4