Image segmentation plays an important role in daily life.Traditional K-means image segmentation is random and easy to fall into local optimization, which greatly reduces the quality of segmentation.In order to overcome these shortcomings, a K-means image segmentation based on Halton sequence improved Manta Ray Foraging Optimization (HMRFO) is proposed.HMRFO uses the Halton sequence to initialize the population to make the individual positions fully uniform, and then introduces Refracted Opposite-Based Learning (ROBL) to improve the global search ability of the algorithm, and finally introduces a new Gaussian mutation strategy to reduce the probability of the algorithm falling into local optimum.Five algorithms are compared in six benchmark test functions, which verifies the effectiveness and feasibility of HMRFO.It is applied to K-means image segmentation and compared with other four algorithms.The results show that HMRFO optimized K-means has better segmentation quality and generalization ability.