Aiming at the high complexity problem of inter-frame coding for screen content video, a fast inter-frame algorithm based on temporal-spatial correlation was proposed. Firstly, the encoding frames were classified to static frames and motion frames according to motion-static detection algorithm. Then, different encoding strategies were used for motion and static frames, respectively. For the static frames, the optimal partition depth and the optimal prediction mode of Coding Unit (CU) were determined based on the CU partition depth and the prediction mode of the temporal correlation. For the static Largest CU (LCU) in the motion frames, the CU partition was terminated prematurely by using the temporal correlation, and the mode selection was only for the large size modes. Whereas for the motion LCU in motion frames, the motion-static characteristic of adjacent LCU was utilized to determine the current CU partition depth and prediction mode. The experimental results show that the proposed method can reduce the average coding time by 46.40% when BDBR is increased by 3.65% compared with the original coding platform. The proposed method can significantly reduce the complexity of screen content inter-frame encoding process in the premise of negligible BDBR (Bjøntegaard Delta Bit Rate) performance loss. Therefore the proposed method is beneficial to the real-time application of screen content video.