JCIC-UIE: Joint Color and Illumination Correction for Underwater Image Enhancement

Abstract

Underwater images often exhibit significant color casts, blurriness, haze, and diminished minute textures because of light attenuation and scattering phenomena. These distortions affect several downstream applications, such as detections and classifications. Despite the potential of deep learning techniques, most of them are computationally intensive and have trouble handling multiple degradations. To address these challenges, we propose Joint Color and Illumination Correction for Underwater Image Enhancement (JCIC-UIE), a novel lightweight encoder-decoder architecture. This framework consists of a novel Jointly Adaptive Chrominance and Illumination Correction (JACIC) module that handles both color and illumination degradation simultaneously in the CIELAB space via an adaptively weighted mechanism. We also incorporate the Shared Multi-DWconv Attention (SMDA) mechanism in each decoder on the encoded feature maps to focus on the most salient features along with capturing the global information. Extensive evaluation on both paired and unpaired datasets shows that JCIC-UIE outperforms existing methods with significantly lower parameters.