The position information of targets is important in the image guidance systems of middle/short-range anti-tank missiles. The encompass pertinence of Markov Random Field and Evolutionary Programming have been applied in many image segmentation problems. Considering the scarce texture and fuzzy edges of infrared, an infrared image segmentation algorithm with simply background is proposed. To quickly segmenting target's areas from an IR image, the algorithm imitates the natural evolutions and uses the competitions of different creatures. The target areas can get higher fitness mark than the background and trivial areas, so that the noise and unimportant areas are restrained. Experimental results have shown that quick segmentation and good restraint to noise can be achieved with the proposed scheme.