Abstract: The two-dimensional minimum error(TME) thresholding method is a viable image segmentation method,but it has high complexity and is hardly used in real-time applications, and it is sensitive to noise, so an improved THE thresholding method was proposed. First the traditional 3*3 template was divided into two complementary parts: acrossing template and a 4-angle template, and the original image was median-filtered with two templates respectively to get two filtered images, then the efficient 2-D histogram was created and the better TME segmentation results were obtwined using the two images,finally the formula of the TME was deduced and simplified to get the simplified formula,and a novel and fast algorithm was deduced with the TME computing features and the formula in order to reduce the computational complexity. Experimental results show that compared with the current THE thresholding algorithm, the proposed method has not only better segmentation performance and robustness, but also its speed is much faster and its memory space is much less.

Key words: Image segmentation, Minimum error thresholding method, Neighborhood template, Recursive algorithm