In human information, visual information accounts for about 60%. Image is the main way for humans to obtain visual information. Image binarization is an important technique in image pre-processing. It has important applications in pattern recognition, optical character recognition, and medical imaging. This paper first introduces the research background, basic principle, elements, algorithm flow, advantages, and disadvantages of bee colony algorithm. To solve the problem that artificial bee colony algorithm is easy to fall into local optimum, this paper proposes an adaptive cauchy mutation artificial bee colony algorithm. This algorithm introduces adaptive factors to expand the search range of bee colony and uses the characteristics of Cauchy distribution to search the global, which improves the universality of bee colony search. Then using stochastic process theory, the adaptive Cauchy mutation artificial bee colony algorithm is analysed theoretically, and the convergence of the algorithm has demonstrated. Finally, we can use Matlab to implement the artificial bee colony algorithm, and we can optimize the Griewank function and Sphere function. In addition, this paper has successfully used in image multi-threshold feature retrieval. Compared with the conventional methods, it shows that the method is more accurate.