Aiming at the complex background suppression problem, a two-prediction-architecture fused complex IR images background suppression algorithm is proposed. In the local region of each pixel, the Nucleus Similarity Degree(NSD) is computed on the basis of the gray value similarity and spatial distribution correlation. Then, the outputs of two spatial filters with different structures are fused as the final background prediction result. Finally, the suppression is done by subtracting the prediction from the original image. The experimental result demonstrates that the proposed algorithm can get better performance in different complex IR images while the structure of the algorithm is simple, and it is suitable for engineering application.