In order to detect dim and small infrared targets,a new approach based on image complex degree is proposed.Four image information entropies of different regions are discussed.Image information entropy describes the average information contents efficiently,but insensitive to point mutations.So information entropy weighted by image variance is introduced to describe image complexity.Two specific analysis models are utilized to obtain image complexity features.It is found that there is much difference in target regions,while almost no difference in non-target regions.After the establishment of the self-adaptive exponent models for the difference image of two complexity features,the dim and small targets can finally be detected with self-adaptive threshold processing.Experimental results show that the proposed method can detect dim and small targets in clouds cluster image with low SCR very efficiently.