MACT: Underwater image color correction via Minimally Attenuated Channel Transfer

Weibo Zhang,Hao Wang,Peng Ren,Weidong Zhang

DOI: https://doi.org/10.1016/j.patrec.2024.11.007

IF: 4.757

2024-11-25

Pattern Recognition Letters

Abstract:Underwater images usually show reduced quality due to the underwater environment where light propagation is affected by scattering and absorption, severely limiting the effectiveness of underwater images in practical applications. To effectively deal with the problem of poor underwater image quality, this paper proposes an innovative Minimally Attenuated Channel Transfer (MACT) method that effectively recovers color distortion and enhances the visibility of underwater images. In underwater images captured from natural scenes, specific color channels are often observed to be severely attenuated. To compensate for the information loss caused by channel attenuation, our color correction method selects the channel with the most minor degradation in the degraded image as the reference channel. Subsequently, we employ the reference channel and the color compensation factor obtained by dual-mean difference to perform adaptive color compensation on different color-degraded channels. Finally, we balance the histogram distribution of the compensated color channels by a linear stretching operation. Extensive experimental results on three benchmark datasets demonstrate that our preprocessing method achieves better performance. The project page is available at https://www.researchgate.net/publication/384252681_2024-MACT .

computer science, artificial intelligence