Abstract: In order to gain better segmentation result of infrared images, and improve the ability to resist noise, an improved 2-D Otsu method was proposed. The inaccurate segmentation in the 2-D gray-neighborhood average histogram was analyzed, and 2-D gray-gradient histogram was adopted. A new algorithm to gain neighborhood average value was put forward. Information of within-cluster was added to amend threshold function, which was further simplified to reduce the calculation complex. Experiments show that the improved method can segment the target better, gain better noise resistance and cost less time.