Abstract: Aiming at multi-focus image fusion, a new image fusion algorithm based on spatial atom structure matching is proposed, it can extract distinctive high frequency components selectively from each input registered image, exploiting matching pursuit signal decomposition. In the result, the fused image has most profound details, without rings or block artifacts. Experimental results show that the algorithm can achieve better fusion effect than the traditional wavelet image fusion method or spatial frequency method, whether from subjective visual effect or objective metric.