Single Underwater Image Restoration Using Variational Framework Guided by Imaging Model With Noise

Abstract

Underwater images typically present poor visibility, color distortion, and noise, which limit the application in several high-level tasks of image analysis. To address these corruptions, a novel method is proposed to reconstruct high-quality underwater images, which is designed by integrating imaging model with noise and variational framework. Specifically, an improved underwater imaging model is first introduced by separating noise from real underwater scene. Subsequently, the hazy curves of degraded colors are decomposed to estimate transmission map, and a color loss prior is employed to correct the transmission map. Moreover, a first-order gradient guided filter is proposed to refine the transmission map. An evaluation formula is designed by combining illumination, contrast, and color deviation priors to accurately search for the background region. Finally, a variational model is established to restore underwater images and suppress noise based on the improved imaging model and image priors. Experimental results validate that the proposed method surpasses several outstanding approaches, demonstrating its well effectiveness in improving contrast, correcting color, and suppressing noise.

Keywords

• Underwater image restoration
• variational framework
• imaging model with noise