The image preprocessing requirements of an infrared image processing system are presented. An Otsu method based on gray gradient mapping function is proposed for the target misclassification occurred easily in the traditional image segmentation algorithm. The method can overcome the shortcoming of the traditional Otsu method based on gray histogram, which can not make use of the local and boundary information in an image. By combining the gradient information with the gray information of pixels, the target can be well separated from the background. The simulation result shows that the algorithm meet the expected specifications and can implement the preprocessing of infrared images.