Photo-Response Non-Uniformity (PRNU) is an inherent feature reflecting the defects of imaging sensors, which can effectively identify the source of cameras shooting digital video. Aiming at the problem of poor recognition effect of network compressed video, a multi-scale neighborhood value shrinkage filtering algorithm based on Stein 's unbiased risk estimation and adaptive edge structure preserving smoothing filtering algorithm was proposed, and a weighted PRNU extraction model was constructed. Firstly, a multi-scale transformation based on the dual-tree complex wavelet was performed on video frames that skip loop filtering. Then multi-scale neighborhood value shrinkage filtering algorithm based on Stein's unbiased risk estimation was used to estimate all high-frequency subbands. After obtaining noise residuals, adaptive edge structure preserving smoothing filtering was used to smooth the complex noise residuals. Then the noise residuals were aggregated using a maximum likelihood estimation method based on quantization parameter weighting to obtain the multiplicative factor of PRNU. Finally, PRNU was obtained through preprocessing. The experimental results on the Vision dataset show that when the video duration is 15 seconds, the AUC values of the proposed model under the moving and rotating reference fingerprints are 0.955 1 and 0.954 9, and the Kappa coefficients are 0.840 3, 0.888 9 and 0.913 2, respectively, which are superior to the existing algorithms.