According to the characteristics of dynamic firing of pulse coupled neural network(PCNN) and the regional configuration of retinal blood vessel network,a new blood vessel tree extraction method based on improved PCNN(IPCNN) is proposed.Firstly,the two-dimensional Gaussian matched filter is used to enhance the retinal image.Then the blood vessel network is segmented by IPCNN.Finally,the vessel tree is obtained via analyzing the region connectivity and using the length filtering.The experiments implemented on the public Hoover database and clinical images indicate that the proposed method outperforms the traditional threshold method and traditional PCNN method on resisting lack-segmentation and over-segmentation,and outperforms the Hoover algorithm on branch vessels extraction and algorithm effectiveness.It is significant for the clinic diagnosis with more practical values.