Abstract: Based on two-dimensional thresholding segmentation algorithm for IR image,the method of 2D maximum classes square error is taken as an example to analyse the shortcoming of the method,then a novel genetic algorithm is presented by combining simple genetic algorithm(GA) and dissipative system theory for segmentation problem.Simulation research on selection of IR image thresholds is carried out for this method.The experimental results demonstrate effectiveness of this method and the comparison results also show at DGA is greatly superior to simple GA.