Systems Engineering and Electronics ›› 2018, Vol. 40 ›› Issue (3): 699-703.doi: 10.3969/j.issn.1001-506X.2018.03.32

Previous Articles     Next Articles

Super-resolution reconstruction based on sparse representation and multicomponent dictionaries learning

XU Zhigang, LI Wenwen, YUAN Feixiang, ZHU Honglei, XU Yamei   

  1. School of Computer and Communication, Lanzhou University of Technology, Lanzhou 730050, China
  • Online:2018-02-26 Published:2018-02-26

PDF (PC)

Abstract:

In order to overcome the disadvantage of some super-resolution methods based on sparse representation that it is not easy to preserve edge and texture detail information and easy to produce visual artifacts, an algorithm based on sparse representation and multicomponent dictionaries learning is proposed. In the stage of dictionary training, the method of morphological component analysis is used to train texture and piecewise smooth dictionaries. In order to extract the detail information of low-resolution images more effectively, the first-order and second-order derivatives are used to extract the piecewise smooth parts. And the texture parts are extracted by the Gabor transform. The L1/2 regularization is used to train the learning dictionaries. In the reconstruction stage, in order to remove blur and minimize visual blocking artifacts, the global constraint term and the nonlocal similarity constraint term are used to optimize the reconstructed high-resolution image. Experimental results can achieve more competitive reconstruction quality of the proposed algorithm both visually and quantitatively.

[an error occurred while processing this directive]