本文以k-means算法为背景，引入信息熵相关知识，从而实现全自动分割图像。然而在利用混合高斯模型对图像进行数据分析时，会发生一定的过拟合现象，导致我们得不到预期的聚类数目。本文设计合理的合并准则，令模型简化，有效地消除过拟合现象，使得最后得到的聚类数目与预期符合。，设计合理的准则改进了图像的全自动分割方法，使得分割结果更加优化(In this paper, k-means algorithm is used as the background, and information entropy related knowledge is introduced to realize full-automatic image segmentation.　However, when the Gaussian mixture model is used to analyze the image data, there will be some over-fitting phenomenon, resulting in that we cannot get the expected number of clusters.　In this paper, a reasonable merging criterion is designed to simplify the model and effectively eliminate the over-fitting phenomenon, so that the final clustering number is in line with the expectation.　A reasonable criterion is designed to improve the automatic image segmentation method and make the segmentation result more optimized.)