Because the transition region thresholding algorithm based on local entropy did not simultaneously consider the amplitude and frequency of the local gray varying of the image, a new transition region thresholding algorithm based on a fusion of local descriptors was proposed. Firstly, in the proposed algorithm, local image feature was extracted by using local image descriptors about local entropy and local variance. Secondly, the feature matrix was constructed by the fusion image feature. Thirdly, the transition region was extracted through opposite feature threshold. Finally, the image was segmented by the grayscale mean of transition pixels. The experimental results show the algorithm performs well in transition region extraction and image segmentation.