Recently,analog visual transmission has attracted considerable attention owing to its graceful performance degradation for various wireless channels.In this study,we propose a novel analog visual communications sys-tem,named DVCast,in which an image denoising algorithm is used to fully utilize spatial correlation;moreover,the variable block size Discrete Cosine Transform(DCT)is used to preserve more correlation information in an image.Obviously,there is a tradeoff between system performance and computing complexity.Therefore,to improve the real-time performance of the proposed system,implementation of Block Matching with 3D filtering(BM3D)and DCT by Graphics Processing Units(GPUs)is introduced.According to DCT block size,i.e.,8 × 8,16 x 16,and 32 × 32,the schemes DVCastg,DVCast16,and DVCast32,respectively,are designed and implemented.Simulations show that DVCast with larger block size achieves better gain and visual quality than reference schemes.Moreover,it requires less computing time.DVCast32 outperforms conventional digital schemes by approximately 3.51 dB and achieves a 1.12 dB gain over state-of-the-art reference schemes.Furthermore,the analysis shows that DVCast can reduce overhead by at least 75％.