Abstract: We proposed an image multiscale geometrical analysis method based on multistage directional guided filter, and applied it to the multisensor images fusion. Firstly, using the guided filter, an approximate subband and some detail subbands at the different scales were obtained by multistage edgepreserving decomposition of the image. Secondly, directional analysis of 
detail subbands was carried out by using wedgeshaped small size directional shear filter, multiple directionalsubbands were generated at the different scales. According to the different physical meanings of approximate subbands and the directional detail subbands, the different fusion criteria were used to merge the decomposed coefficients respectively. Finally, the reconstructed image was obtained by simple superposition calculation of the merged coefficients. This method effectively solved the problems that filters lacked selfadaptability and image analysis did not have translation invariance in the other multiscale decomposition. The experimental results of multigroup image fusion show that the performance of this method is better than other typical multiscale decomposition methods in the process of image fusion.

Key words: image fusion, guided filter, edgepreserving, shear filter, fusion criterion