Abstract: An improved PRNU noise preprocessing algorithm is proposed in this paper to address the problem of losing part of the sensor pattern noise information during suppressing non-unique artifacts. The noise residual obtained by filtering was processed to remove diagonal artifacts. The noise spectrum was processed by iterative real-time least squares smoothing based on additive half-quadratic optimization. The pattern noise information lost after smoothing was cyclically reprocessed and recycled. The smoothing items were accumulated to obtain the preprocessed noise. The algorithm was evaluated on the public dataset Dresden. The results show that it outperforms existing algorithms in the performance of source camera identification for JPEG-compressed images.