Visibility Improvement in Underwater and Traffic Images Using Color Balance and Fusion

Abstract

We present a robust method for enhancing underwater images that suffer from visual degradation due to light absorption and scattering in water. The approach works with a single image and does not rely on additional hardware or prior information about the underwater scene or conditions. The process involves generating two distinct versions of the image from a white-balanced and color-corrected base, each emphasizing different visual features. These images are then fused using spatial weight maps that are specifically designed to enhance color contrast and preserve edge information. To minimize the introduction of low-frequency artifacts caused by abrupt transitions in the weight maps, a multiscale fusion technique is applied. Our method has been extensively tested through both visual and numerical evaluations, demonstrating improved visibility in darker regions, better overall contrast, and enhanced detail sharpness. The technique also performs consistently across different camera settings and has been shown to improve the performance of image processing tasks like segmentation and feature matching.