基于分形的改进Grabcut目标自动分割

针对传统Grabcut 算法依赖人工互动，分割效率低的缺点，提出一种基于微分计盒快速算法的Differential Box-Counting Grabcut（DBC-Grabcut）方法，实现复杂自然背景下的人工目标自动分割。应用微分计盒快速算法检测出人工目标的轮廓，进而确定出Grabcut的初始分割轮廓和混合高斯模型参数，运用最小割方法分割图像，再通过少量的迭代使能量函数最小化完成目标分割。实验结果表明，该算法能够自动分割出人工目标，且分割结果较为完整，能充分保留目标的原始信息。

Automatic target segmentation by improved Grabcut based on fractal

In view of the disadvantages of traditional Grabcut algorithm rely on human interaction with low efficiency in segmentation, an improved Grabcut algorithm based on the fast differential box-counting method is proposed. It can segment the artificial targets under the complex natural background autonomously. The contour of the artificial targets is detected through fast differential box-counting method, then the initial rectangle and the parameters of Gaussian Mixture Model needed in the Grabcut can be obtained. Utilizing the min cut method to segment the pixels of image, and the target is segmented by energy minimization through several iterations. Experimental results show that the proposed algorithm can segment the artificial target relatively complete under the complex natural background, and can reserve the target’s original information sufficiently.