Consistent Coding Scheme for Single-Image Super-Resolution Via Independent Dictionaries

Wenming Yang, Yapeng Tian, Fei Zhou, Qingmin Liao, Hai Chen, and Chenglin Zheng


In this paper, we present a unified frame based on collaborative representation (CR) for single-image super-resolution (SR), which learns low-resolution (LR) and high-resolution (HR) dictionaries independently in the training stage and adopts a consistent coding scheme (CCS) to guarantee the prediction accuracy of HR coding coefficients during SR reconstruction. The independent LR and HR dictionaries are learned based on CR with l2 -norm regularization, which can well describe the corresponding LR and HR patch space, respectively. Furthermore, a mapping function is learned to map LR coding coefficients onto the corresponding HR coding coefficients. Propagation filtering can achieve smoothing over an image while preserving image context like edges or textural regions. Moreover, to preserve the edge structures of a super-resolved image and suppress artifacts, a propagation filtering-based constraint and image nonlocal self-similarity regularization are introduced into the SR reconstruction framework. Experimental comparison with state-of-the-art single image SR algorithms validates the effectiveness of proposed approach.


Fig 1. Reconstruction framework of proposed method.


Fig 2. Reconstructed results of Butterfly by different SR methods with an upscaling factor of 3.

  • Sorry
    • This work is cooperated with a company, so the original source code of the paper is not available.
  • But
    • We would be happy to assist you in making comparison with our method. (Please send me your test images!)

  • J. Yang, J. Wright, T. S. Huang, and Y. Ma, “Image super-resolution  
  • via sparse representation,” IEEE Trans. Image Process., vol. 19, no. 11, 
  • pp. 2861–2873, Nov. 2010.
  • S. Wang, D. Zhang, Y. Liang, and Q. Pan, “Semi-coupled dictionary learning with applications to image super-resolution and photo-sketch synthesis,” in Proc. IEEE Conf. Comput. Vis. Pattern Recog., Jun. 2012, pp. 2216–2223.
  • K. Zhang, D. Tao, X. Gao, X. Li, and Z. Xiong, “Learning multiple linear mappings for efficient single image super-resolution,” IEEE Trans. Image Process., vol. 24, no. 3, pp. 846–861, Mar. 2015.
  • D. Weisheng, Z. Lei, S. Guangming, and L. Xin, “Nonlocally centralized sparse representation for image restoration,” IEEE Trans. Image Process., vol. 22, no. 4, pp. 1620–1630, Apr. 2013.
  • C. Dong, C. L. Chen, K. He, and X. Tang, “Learning a deep convolutional network for image super-resolution,” in Proc. Eur. Conf. Comput. Vis., 2014, pp. 184–199.
  • C. Dong, C. Loy, K. He, and X. Tang, “Image super-resolution using deep convolutional networks,” IEEE Trans. Pattern Anal.Mach. Intell., vol. 38, no. 2, pp. 295–307, Feb. 1 2016.