Realizing th at orthogonal wavelet coefficients which are designed to decouple the correlation between pixels and the retained coeffic ients can not provide enough information for the missing ones,a new curvature driven diffusion (CDD) image inpainting model based on three channel multiwavelet tight frame is proposed in this paper.Considering that in the original CDD inpainting model, the isophotes are still approximate by staright lines,a new model is proposed.In the new model,image inpainting is performed according to improved curvature fu nctions in the normal while information contained in the observed region can permeate into dam aged area in the tangential.The model presented this paper combines a novel multiwavelet tight frame decomposition technique with a improve d three order CDD inpainting model.Multiwavelet tight frames are redundant,so information contained in th e available coefficients will provide information for the missing coefficients.The proposed model has effecti ve control on features and sharp edges of inpainting images,even with massive loss of coefficie nts .Numerical experiment by using Split Bregman on various loss scenarios and natural images shows the effectivene ss of the proposed approach.