Considering the OCT image noise characteristics and the general structure of the saddle point, this paper proposes a primal-dual algorithm for OCT image noise reduction. After incorporating the log of the speckle noise model, the primal-dual structure is used, which is easy to implement. From our experiments, this algorithm is compared with the mean filter and semisoft shrinkage methods, and it shows that noises in OCT images were removed and important details were preserved simultaneously. The proposed algorithm reduced the speckle noise in OCT images effectively and improved the quality of the images. At the same time, the wide applicability of the proposed algorithm was demonstrated on image segmentation, target recognition and motion estimation.