In order to effectively eliminate noise image with impulse noise, Gaussian noise and even mixed noise, the work improves the 3D block-matching algorithm and puts forward a new image denoising algorithm. Firstly, the 3D matrix was constructed with the similarity between image block of noise image. Then, the noise was attenuated by hard-thresholding between the image blocks, and the initially estimated denoised images were obtained through the weighted average reconstruction of image blocks. Finally, block-matching was performed on the initially estimated denoised images, and Wiener filtering and weighted median filtering were conducted in and among image blocks, finally denoised images were obtained. Simulation results showed that the proposed algorithm had ideal denoising effect on common noise of image acquisition, and the PSNR value was more than 31 dB. Compared with Wiener filtering, median filtering and hard threshold wavelet filtering, PSNR results of Gaussian noise, impulse noise and mixed noise with this algorithm were 31.5334～36.6466 dB, higher than other algorithms. The highest difference value reached 12.08 dB. In conclusion, image denoising method combined with median filter and 3D block-matching algorithm can better reduce a variety of noises and is an excellent denoising algorithm.