Based on the nonsubsampled contourlet transform (NSCT) and two denoising models (i.e., fractional power model and
cross-scale correlation model), an efficient pre-processing algorithm for infrared image is proposed. In our algorithm, the
NSCT is used to decompose the image at different scale and orientation, and then implement pre-processing in the
frequency domain, at last reconstruct coefficients to obtain ideal infrared image. The key of the proposed algorithm is
pre-processing which includes noise removal and information enhancement. To reduce the two kinds of noises (i.e.,
Gaussian noise and shot noise) efficiently, the two models referred are applied to the NSCT coefficients respectively.
The filtered results are fused to learn from the strong points of each denoising methods to offset the weakness of each
other. Later, the denoised coefficients are classified to edges and noise and modified by a nonlinear mapping function.
Experiments carried on infrared images show that the new algorithm can reduce the Gaussian noise and shot noise
efficiently, while keeping the detail information well. Both in the objective performance index and subjective viewing
assessment, the new algorithm is superior to the DWT-based method as well as the traditional method.