针对红外图像有别于一般灰度图像的特点，常用的灰度级-平均灰度级2维直方图区域划分在红外图像分割中效果不佳，为此提出一种改进的灰度级-梯度2维Otsu阈值选取方法，选取合适的梯度算子，利用改进的粒子群优化算法寻找分割阈值，在算法中加入有效判断早熟停滞的方法，一旦检索到早熟迹象，便随机改变最优解的任意1维分量值，使其跳出局部最大，实现全局寻优过程的快速收敛。仿真实验结果表明，该算法由于使用新的2维直方图，分割后的红外图像边界形状准确，特征细节清晰，运算速度也得到了有效提高。

To consider the differences between infrared image and common gray level image, the effects of regional division are not good when the commonly used gray level-average gray level 2D histogram is used in infrared image segmentation. An improved Otsu threshold selection method based on gray level-gradient 2D histogram is proposed in this paper. The appropriate gradient operator is chosen. An improved particle swarm optimization(PSO) algorithm is used to search for the segmentation threshold. An effective method that identifies premature stagnation is embedded to PSO, so once premature stagnation happens, a randomized solution, as a substitute for current optimum, is used for particles to go out of the local optima. The simulation experiments demonstrate that the algorithm proposed in this paper achieves accurate borders and clear details of features after infrared image threshold because of the new 2D histogram. The compute speed is improved effectively.