A novel approach using some information to define the regions of interest is proposed for combining multisensor images. Firstly, the original images are separated into object region and background region. Then based on using the à trous wavelet transform to decompose the original images, the fused coefficients are obtained by performing different fusion process to object region, background region and region boundaries between object region and background region. Finally, the fused image can be constructed by applying the inverse transformation. Experiments results show that the proposed approach outperforms the pixel-based and window-based image fusion methods.