Study to the influence of registration error to fusion image quality

Image registration is the precondition to realize the image fusion course. Based on the analysis to registration error study the quality change of fusion image under different translating shift. First, choose the principal component analysis method, Laplacian pyramid method and shift invariant discrete wavelet transform method as the fusion algorithm to get different fusion image. Then, select the evaluation method such as entropy, mutual information and structure similarity as the assessment metrics. With the different translation modification between input images, experimental result show it is differ from established conclusion that registration precision must be restricted in the bound of one pixel or sub-pixel. Different fusion algorithms have the capacity to overcome the influence of slight translating error. The fusion image quality is degraded slowly in a certain extent, which still can be accepted by a great number of observers. Otherwise, similarity structure assessment index show also the excellent performance to different fusion images because it is preferable to apply locally on image blocks rather than the entire image.