The extraction method of urban trees based on multisource remote sensing data is of great significance for urban resource investigation, health status evaluation, and scientific management. To further improve the accuracy of tree extraction, this paper combined the advantages of LiDAR and orthophoto data and proposed a tree extraction algorithm based on automatic feature segmentation. This algorithm is used to identify and extract the shadowed regions; the correlation between normalized difference vegetation index (NDVI) and digital surface models (DSM) local entropy features is used to reduce the background extraction threshold by combining with histogram subtraction. Experimental results show that the proposed algorithm, which has been verified using the ISPRS Vaihingen dataset, has a high precision of tree extraction on multiple datasets and is robust to a certain degree, thus being suitable to be used in complex environments.