Abstract: We prove that linear optimal fusion method is equivalent to optimal fusion,thus reduce the calculation complexity greatly.Furthermore,we extend the algorithm by introducing weighted average of corresponding pixels on the multi-spectral images to emphasize the spectral characteristics of the interested region of each source image.An adaptive method for determining the weight of each pixel is described.For images with weak correlation,we adopt a grouping fusion approach.Experiments show our algorithm is effective and can meet the requirements of different applications.