A number of image filtering algorithms based on nonlocal means have been proposed in recent years which take advantage of the high degree of redundancy of any natural image. The block-matching with 3D transform domain collaborative filtering (BM3D) proposed in [1] achieves excellent performance in image denoising. But the choice of shrinkage operator in block-matching step is not discussed, only given the threshold by experience in its related papers. In this work, we introduce an improved version of BM3D with adaptive block-match thresholds. The proposed method firstly seeks the relationship between the Structural Similarity index (SSIM) [2] and match distance in blocks and obtains the data with fine SSIM values. Then, compute the Noise level and Gradient values in blocks of the same block size. Finally, surface fitting is adopted to get a formula which applies weak thresholds for flat blocks and strong thresholds for detail blocks. Experiment results are given to demonstrate the same class of denoising performance with less time-consuming to slightly noisy image and good improvement in denoising performance to seriously noisy image.