In the process of billet detection and recognition, how to accurately divide the characters at ends of steel in the complex scene is a highly complicated intelligence problem. In order to solve this complex problem, a segmentation algorithm based on intelligent multi-agent is proposed in this paper. The algorithm takes these functions: the segmentation of characters, the combination of regions, the division of regions and the calculation of features, as the agents in the subordinate control layer. And then, these agents work in coordination with each other in the control of the master agent. Moreover, the segmentation information is fed back to the master agent to control and analyze the agents. At last, thebillet characters are divided accurately through the proposed algorithm of intelligent multi-agent. The segmentation experimental results show that the proposed algorithm divides the billet characters in the complex scene accurately and steadily. What’s more, the algorithm solves the difficult problem of the accurate segmentation of billet characters in the complex scene, and provides the guarantee for the characters recognition.