Single-image water surface reflection and shadow removal via gradient-color guided region-growing segmentation and inpainting

Reflections and shadows on water surfaces hinder the extraction of valuable information and make computer algorithms produce imperfect results, yet research on effectively removing them from water surfaces remains scarce. In this paper, we present an innovative algorithm that simultaneously removes both reflections and shadows from a single image. The algorithm begins by lowering the overly high V components of pixels in the HSV space to preliminarily remove reflections. Then a novel gradient-color guided region-growing segmentation and inpainting method is applied. This method defines a local cost that integrates gradient magnitude, gradient direction, and color information, enabling precise segmentation and inpainting of reflections and shadows on water surfaces. Experimental results demonstrate the algorithm’s effectiveness in complex scenarios. This approach not only removes reflections and shadows but also shows potential for applications like riverbank line identification, making it valuable for downstream tasks such as recognition and tracking.