In order to reduce the storage and transmission cost of video, lossy compression is in frequent use, which however would incur various types of artifacts in the video and affect users’ subjective visual experience. The single frame method cannot be directly applied to video processing, because they independently process each video frame, limiting the use of spatial information and causing limited effectiveness. Inter-frame alignment or temporal structure was often adopted in multi-frame methods to enhance the reconstruction results by utilizing the temporal information, but there remains much room for improvement in alignment performance. To solve the above problems, a multi-frame spatio-temporal compression artifact removal method was proposed to achieve better alignment fusion through the alignment fusion design. This method efficiently utilized the multi-frame spatio-temporal information to reconstruct high quality videos. The experimental results show that the proposed method outperforms other comparative methods on a number of test compressed videos with different resolutions (HM16.5 under low delay P), and that it can achieve an average improvement of 0.13 dB on the objective index peak signal to noise ratio (PSNR) compared with the state-of-the-art multi-frame method STDF. Meanwhile, the proposed method can yield promising results in subjective comparisons, reconstructing a clearer picture with a good effect of compression artifact removal.

compressed image restoration, block effect removal, video enhancement, multi-frame alignment fusion, deformable convolution