A multi-level residual reconstruction based image compressed sensing recovery scheme

Abstract

Traditional image compressed sensing recovery focuses on the research of sparse representation and reweight processing of measurements. However, the image content and structural feature vary dramatically in different natural images. The improvements in reconstruction quality brought by exploring new sparse representation models are not satisfactory. This paper focuses on a multimedia application scenario where the encoder is resource-constrained and the decoder has powerful computing ability. A novel image compressed sensing recovery scheme based on multi-level residual reconstruction is proposed to further improve the reconstruction quality. By converting the original image recovery to the multi-level residual image recovery, the reconstruction process is divided into three phases. The hidden information in image recovery is fully utilized. Moreover, a constraint-adaptive recovery model is proposed to perform the initial reconstruction of the original image and the initial residual image reconstruction. Combining the multihypothesis prediction, the final recovered residual image is obtained in the secondary residual image recovery phase. The final recovered image is obtained by combining the recovered original image and the residual image. Experimental results show that our proposal outperforms the state-of-the-art methods for image compressed sensing reconstruction in both objective and subjective quality.