In view of the problems of the current thresholding method based on 2-D minimum cross entropy, such as computing complexity, an improved image segmentation method based on 2-D minimum cross entropy is presented. Firstly, a neighborhood mask was selected according to the image type including the noise and a corresponding 2-D histogram was created to improve the segmentation performance. Then, the thresholding formula of 2-D minimum cross entropy was simplified, and through the defined array operations, the recursive algorithm was combined into a new search algorithm. The gray limits of the image and the neighborhood image were obtained, and in the limits, the new recursive algorithm was used to search the best threshold vector and to reduce the computing complexity. Finally, the neighborhood image was segmented with the key threshold to obtain better segmentation effects. Experimental results show that the proposed method’s segmentation effect and its anti-noise are better than those of the current thresholding method based on 2-D minimum cross entropy, and its computing time is much less, below 0.05 second.