Abstract: A novel passive millimeter wave image restoration method is proposed,which aims to overcome the shortcoming that Fourier and wavelet domain regularization methods can not de-noise effectively and maintain target features simultaneously.The new method takes advantage of sparse representation's merit of representing signals flexibly.It learns from the millimeter wave image after inverse filtering by using K-clustering with singular value decomposition (K-SVD) algorithm to obtain basis functions adaptively for image restoration.Comparing with Fourier and wavelet domain regularization methods, the proposed method employs an adaptive method.So it can maintain target features better and de-noise more effectively, which leads to better image restoration.When the method was used in the restoration of simulated passive millimeter image, good result has been obtained.Therefore, it is an effective passive millimeter imaging method.