AFM image is often distorted due to the influence of the tip shape. This problem has become one of the major obstacles in nano-observation. Therefore it is necessary to develop a compensation method to improve the quality of the AFM image. The common used method today is image reconstruction with deconvolution operator based on the tip shape built through blind tip estimation algorithm. However, the current algorithm has difficulties in getting the optimal noise threshold, which is essential for reducing the noise influence during the tip evaluation. This paper proposed a new method to solve this problem, through which the optimal threshold can be easily determined based on the comparison of tip section profile P-alpha and tip standard profile P-s. The experimental results validate the effect of the proposed method, and it also shows that the image quality has been greatly improved.