Objective Retinal blood vessel extraction is of great importance for the early diagnosis. Aiming at the issues that the segmentation accuracy is not high in the majority of retinal blood vessel extraction algorithms, a new method is introduced. Methods Firstly, the vessel information is extracted by using top-hat transformation. The morphological ‘top-hat transformation’ operation is conducted by using the ‘disc’ structuring element to smooth the image background and highlight the blood vessels. Secondly, Otsu’s segmentation is used to get its binary image. Thirdly, according to the structural information and Geometric features of retinal blood vessel, the false blood vessels are removed by measuring the connected domain. Finally, in order to maintain the continuity of blood vessels, we use morphological dilation for blood vessels to connect broken blood vessels and reduce the experimental error. Results Through the above steps, the automatic identification of vessel is realized. Conclusions The experimental results indicate that the proposed algorithm can effectively detect the blood vessels of fundus image with high segmentation accuracy.