Ghost imaging (GI) is an important technique in the fields of quantum imaging and classical optical imaging, and it can solve the problems which are difficult to solve by the traditional imaging techniques in the optically harsh environments. In this paper, we present the iterative denoising of GI based on an adaptive threshold method. This method is abbreviated as IDGI-AT, which takes the advantages of adaptive threshold, differential, binarization and iterative operation method, and can enhance image quality in GI. In addition, this method can reduce the number of measurements. As is well known, the enormous number of measurements and poor reconstruction quality are obstacles to the engineering application of GI. The correlation noise leads to low signal-to-noise ratio and low imaging efficiency in GI as well. Therefore, we establish a denoising model, which can reduce correlation noise and improve reconstruction quality. We first analyze the iterative denoising of ghost imaging (IDGI) theory, and use the adaptive threshold technique to calculate the ideal threshold associated with the correlation noise. It should be noted that the threshold can be obtained by this method under the condition without requiring prior knowledge of the object. Afterwards, we can construct the correlation noise in this denoising model. In the IDGI, the differential ghost imaging (DGI) image is taken as the initial iteration value. We use the adaptive threshold method, which is different from IDGI, to binarize the initial value of each iteration to make it closer to the original object's transmission coefficient. After three iterations, we can obtain a higher-quality reconstruction image. In order to demonstrate that the IDGI-AT is available, a GI experimental system with a pseudo-thermal light source is set up. The considerable simulation and experimental results show the advantage of our scheme in terms of removing reconstruction image background noise. Especially, the visual effects and peak signal-to-noise ratio values are improved in comparison with those from the traditional GI, DGI and IDGI. Besides, we demonstrate the role of binarization in our scheme. For a binary object, the iterative value binarization can achieve better image quality than that in the case without binarizing the iterative initial value. Therefore, this novel method is likely to provide an alternative mean for GI and further pave the way for the application fields of GI, such as lidar, biomedical engineering, etc.