Aiming at the image segmentation problem in highway crack detection, using the method of machine learning field, taking crack image as the research object, after analyzing the classic K-Means clustering algorithm and Gaussian Mixture Model (GMM), this paper proposes a method of segmenting highway crack images using global K-Means and Gaussian mixture models. Based on the clustering of global K-Means algorithm, this method adopts an improved strategy. At the same time, each probability distribution of traditional GMM is further subdivided into individual probability components. In the process of calculating auxiliary variables, EM algorithm is adopted. The experimental results of the simulated image and the actual crack image show that the proposed method has higher accuracy and better stability than the segmentation method using the traditional K-Means algorithm and the ordinary GMM.