According to the problem that regularized orthogonal matching pursuit (ROMP) nee ds to forecast sparsity level in compressed sensing,and would be unstable if the sparsi ty level is improperly estimated,the improved ROMP is proposed.Because the observation signal can inherit the original signal,an adaptive weak selection criterion is introduced in the process of selecting candidate atoms.A weak selection strategy is set by the information o f observation signal,then it adjusts the sparsity adaptively.The algorithm is applied in the medical image fusion under compressed sensing,then a fusion rule based on the similarity of observation signal struct ure is proposed.Taking the weight of observation signal information as fusion rule,when the structural sim ilarity between the observed signals is higher,it shows that the original signals have the sam e similarity.In the same way,the larger information of observation signal is chosen as fused observation signal when the structural similarity between the observed signals is lower.The experiment al results show that the reconstructed image quality of the improved ROMP algorithm is better than those of OMP,ROMP,SAMP,and the peak signal to noise ratio is improved by 2~3dB.When applied to medical image f usion,the fused image has good human visual effect and retains most of the feature information in the source image,and further it can get high quality fusion results in a short time.