结合互信息熵差测度的高斯混合模型图像分割

A Gauss Mixture Model with Difference of Mutual Information for Image Segmentation

从分割图像与原图像的内在联系出发,提出了一种基于高斯混合模型与互信息熵差结合的分割算法———GMM-DMI算法.利用期望极值化方法确定高斯混合模型的各分量参数,以互信息熵差为模型选择准则,计算前分割图像与当前分割图像的互信息熵差,互信息熵差达到最小时即为最优解.实验结果表明,本算法所得到的目标图像的区域保持形状且定位性能好.

According to the internal relations between original image and segmented image, the gauss miture model is combined with the difference of mutual information （DMI）. The parameters of GMM can be obtained by using Expectation Maximization method, and in iteration process, an optimal component number will be determined by minimizing the DMI between the previous and current segmented images. The experimental results indicate that the proposed method has not only visually better segmentation effect but also better localization property.