A real-time underwater image enhancement method based on red channel correction and guided-filter-refined DCP
Abstract
Remotely operated vehicles images taken under water suffer color distortion, poor contrast and blurred details as well as decreased visibility because of wavelength-dependent light absorption and scattering. The degradations cause serious damage to future tasks involving visual perception. In order to overcome such problems, this paper suggests a real-time underwater image enhancement approach relying on red channel correction with guided-filter-based dark channel prior. Precisely, adaptive red channel compensation is applied before estimating the dark channel to reduce the negative impact of extreme red-light attenuation on the estimation of the background light and transmission. Then, the guided filter is used to refine the coarse transmission map at a lower computational cost without compromising major edge structures. Image restoration and local contrast enhancement follow to create the final enhanced output. Experimental evaluations on the U45 dataset demonstrate that the proposed method achieves an average underwater image quality measure of 4.5200, an average underwater color image quality evaluation of 0.7665, and an average processing time of 0.0186 s per frame under the reported settings. These findings substantiate the method's efficacy as a practical preprocessing solution for real-time underwater vision systems.