<p>The arithmetic optimization algorithm (AOA) is a method with good optimization ability. However, when AOA is dealing with image segmentation, its weak search ability, low convergence accuracy and slow convergence speed are still its main shortcomings. In order to better solve multi-threshold image segmentation based on symmetric cross-entropy, an improved arithmetic optimization algorithm based on neighbor information sharing (NAOA) is proposed. Firstly, a neighborhood dimension learning-based hunting (NDLH) search strategy is introduced to enhance the searching ability of the algorithm. Secondly, the neighbor gaze cue learning strategy (NGCL) is used to improve convergence accuracy. Finally, the Lévy flight strategy is adopted to accelerate the convergence speed. In addition, we introduce a new boundary constraint method that can effectively use population information. The integration of these strategies into AOA enables effective information sharing among population individuals and improves the overall search efficiency of the algorithm. In order to evaluate the convergence performance of NAOA and the feasibility and effectiveness of solving practical problems, NAOA is applied to CEC2017 test functions and symmetric cross-entropy multi-threshold image segmentation. Experimental results of NAOA are compared with those of other optimization algorithms, and nonparametric statistical analysis of experimental results is performed using the Wilcoxon rank-sum test. Simulation results and analysis show that NAOA has better convergence accuracy and speed than other algorithms in most cases.</p>