To reconstruct high-resolution(HR) images from a sequence of low-resolution(LR) compressed images,this paper proposes a novel algorithm focused on super-resolution reconstruction of H.264 compressed video,which is based on the integer DCT transform-domain quantization noise.Firstly,models of compressed video and integer DCT transform-domain quantization noise are surveyed.Then the reconstruction algorithm under the POCS theory is proposed.Experimental results demonstrate that this algorithm has a great improvement in subjective visual quality and low computation complexity,in which PSNR can reach 30 dB and iterations are less than 5 times.