For getting more real Infrared images, a new method of retrieving Infrared images using low-level images is proposed. Firstly, low-level images with different material properties are classified according to ground material properties using improved K-means clustering algorithm. And then, the total radiant energy into simulation Infrared thermal imaging system in special bands is calculated. And simulation Infrared images are obtained through simulating sensor imaging principle. The generation of Infrared scene sequence images in some area is finished with the algorithm and real measurement environment meteorological parameters. Experimental results show that the sequence images generated with the method basically accord with the change rules of real Infrared images, so an effective way is provided for low-level images simulation generation.