Minimax Concave Penalty Regression for Superresolution Image Reconstruction

Xingran LIAO, Xuekai WEI, Mingliang ZHOU*

*Corresponding author for this work

Research output: Journal PublicationsJournal Article (refereed)peer-review


Fast and robust superresolution image reconstruction techniques can be beneficial in improving the safety and reliability of various consumer electronics applications. The least absolute shrinkage and selection operator (LASSO) penalty is widely used in sparse coding-based superresolution image reconstruction (SCSR) tasks. However, the performance of the previously developed models is constrained by bias generated by the LASSO penalty. Meanwhile, no efficient and fast computing algorithms are available for unbiased l0 regression, and this situation restricts the practical application of l0-based SCSR methods. To address bias and efficiency problems, we propose a model called minimax concave penalty-based superresolution (MCPSR). First, we introduce a minimax concave penalty (MCP) into the SCSR task to eliminate bias. Second, we design a convergent, efficient algorithm for solving the MCPSR model and present a strict convergence analysis. Numerical experiments show that this model and the designed supporting algorithm can produce reconstructed images with richer textures at a fast computing speed. Moreover, MCPSR even shows robustness in the superresolution reconstruction of noisy images compared with other SCSR methods and has two flexible parameters to control the smoothness of the final reconstruction results.

Original languageEnglish
Pages (from-to)2999-3007
Number of pages9
JournalIEEE Transactions on Consumer Electronics
Issue number1
Publication statusPublished - Feb 2024
Externally publishedYes

Bibliographical note

Publisher Copyright:
© 1975-2011 IEEE.


  • fast computing method
  • Image superresolution
  • minimax concave penalty regression
  • sparse coding


Dive into the research topics of 'Minimax Concave Penalty Regression for Superresolution Image Reconstruction'. Together they form a unique fingerprint.

Cite this