Block matching 3D filtering (BM3D) algorithm is more effective than traditional denoising methods especially for Gaussian noise. However, the traditional hard-threshold used in BM3D algorithm can not recognize the noise intensity in the process of removing additive noise with BM3D, some image details will be lost. Aiming at this problem, an improved BM3D algorithm is proposed. Firstly, the traditional hard-thresholding of the BM3D method is substituted by an adaptive filtering technique. This technique has a high capacity to acclimate and can change according to the noise intensity. Secondly, when the noise intensity is less than the threshold, the new TV model is used to replace the hard-threshold filtering, when the noise intensity exceeds the threshold, the hard and soft thresholding algorithm is used to replace the hard-threshold filtering. Through the adaptive threshold, the high noise and low noise image feature areas are screened out, targeted denoising is carried out, and the edge details are preserved. Experimental results show that the performance of the improved BM3D method is better than traditional methods.