The effectiveness of one de-noising method for different noise images is different, a method based on wavelet domain for identifying image noise types was used. The histogram of wavelet HH sub-band’s coefficients was used to identify two types of main noise: Gaussian white noise and salt & pepper noise. After analyzing the infrared image series, it is found that the long-wave infrared images are mainly affected by the Gaussian white noise, the medium-wave infrared images are mainly affected by the salt & pepper noise. An improved wavelet global threshold de-noising method is reported. The simulation experiments indicate that the PSNR of this method is increased by 2～3 dB, processing speed is more faster, compared with traditional methods.