Effective underwater image restoration based on color channel compensation
Horia Baltă, Codruta Orniana Ancuti, Cosmin Ancuți · OCEANS 2021: San Diego – Porto · 2021
In this work we introduce an effective underwater restoration technique built on the Dark Channel Prior (DCP). However, different than the previous solutions, we built on the statistical observation that in underwater images the vanishing of a color channel tends to move to the the opponent color away from the origin. As a result, we first perform a color compensation step, which consists in considering that the spatial average of colors in a natural scene is neutral gray. After employing this color-correction step, we estimate the transmission map from the weakest color value over a neighborhood and the backscattered light value by searching for the brightest location of the image. Finally, in order to compute the restored image, we simply invert the optical model equation using these two estimated values. The experimental results prove that our approach is competitive, yielding comparable and even better results than compared with the recent specialized underwater techniques.