Abstract: Infrared ship segmentation is very important for automatic infrared ship recognition in the sea. The thresholding based algorithms are widely applied to segmentation due to the intrinsic merits. The threshold is set based on the assumption that the intensity relation between target and background is known, but the assumption is incorrect in the medium wasve infrared images. Because of the sensitivity to the environment, the target in the medium wave infrared images may be bilateral polarity, so the adaptive thresholding can not be realized. Considering the adaptive thresholding ability for segmentation of bilateral polarity ship target, a new maximum two dimensions entropy segmentation algorithm is proposed. The multi-scale local variance-entropy variety and variance of gradient direction are used to build the two dimensions histogram, the optimized threshold vector are obtained by maximizing two dimensions entropy using the particle swarm optimization algorithm. Then the fine segmentation is performed by iterative thresholding on the coarse segmentation results to get the accurate segmentation result. Experimental results indicate that the proposed algorithm can get good performance in bilateral polarity target segmentation.