The inherent shortcoming of POCS (Projection Onto Convex Sets) is its sensitiveness to noise. The restoration quality of POCS based super resolution will severely decline when the noise is larger. In practical applications, the low resolution images generally include some kinds of noise, such as camera internal noise, transmission system noise and coherent noise. Therefore POCS cannot be used directly in super-resolution restoration for observed low resolution images. In order to solve the noise sensitive problem of the traditional POCS restoration algorithm, we firstly propose to optimize the BM3D (Block-Matching 3D) filtering by mean pre-screening of image blocks and limiting the number of image blocks. Then we combine the optimized BM3D filtering with POCS restoration in this paper. Experimental results show that the proposed POCS super resolution restoration algorithm based on BM3D can achieve better restoration effect when the low resolution images contain noise. Furthermore no noise can be perceived in the restored high resolution image basically.