On the basis of traditional anisotropic diffusion filtering algorithm, an adaptive gradient threshold based anisotropic filtering algorithm is proposed to suppress the infrared complex background, filter out the noise effectively and enhance the infrared dim target. According to the local features of a image and its gradient features in different directions , the algorithm can determine the point of a pixel to be a noise or an image, and can point out the point of the pixel to be in the smooth region or the edge region. Based on above, it proposes a method to calculate the gradient threshold (K value) of the edge function adaptively and to solve the shortcoming that the K value is fixed and single in the edge function of the traditionall anisotropic filtering algorithm. This experiment shows that this algorithm increases the denoising function, suppresses complex background effectively, and enhances Signal to Noise Radio(SNR) of the image by 2 times as compared with that of the original anisotropic filtering algorithm and other background suppression algorithms.