Abstract: An improved clustering segmentation algorithm based on Gaussian mixture model and K-means is proposed in this paper to fix the problem of difficult to extract tooth target region in oral CT images with blurred boundary and low contrast. First of all, the image is preprocessed by anisotropic filtering to achieve de-noising and smoothing while enhancing details of the image. Secondly, K-means is used to complete a kind of initial division, and initial values of EM algorithm iteration is then given according to the pixel value after classification, so as to speed up the algorithm iteration to an optimal solution. This greatly reduces the number of algorithm iterations and effectively fixes the problem of GMM caused by randomly selecting the initial value points when EM algorithm solves parameters. Finally, we use EM algorithm to learn GMM and complete ML segmentation. Our experimental results indicate that the improved method reduces the computational complexity, has strong robustness to noise, and can obtain more ideal segmentation results.

Key words: oral CT image, anisotropic filtering, K-means algorithm, Gaussian mixture model, robustness