Most natural images are texture images, so compared with the traditional color features, the texture images have the characteristics of the descriptive directionality and scale difference. GrabCut image segmentation is characterized by semi-interaction, and the segmentation of the image foreground and background is the adaptive decision of the final segmentation termination process based on establishing Gaussian mixture model(GMM) probability density distribution to achieve the description of similarity, combining the max-flow min-cut image segmentation with GMM parameters of iterative update foreground and background, and utilizing Kullback-Leibler(KL) criterion of Gaussian mixture probability distribution of foreground and background. The experiment results show that the proposed method has the ideal segmentation effect on the synthetic texture images and natural texture images.