Evaluation of Laser Image Enhancement and Restoration for Underwater Object Recognition

Underwater object recognition is a challenging task due to the degradation of image quality caused by the scattering of light in the water medium. A variety of techniques for image enhancement (IE) and image restoration (IR) have been proposed over the years; however, these methods have been developed primarily for imaging of scenes with RGB cameras, not object recognition with laser images, which is the focus of this work. Due to the different radiometric transfer properties of air and water, methods developed for airborne applications are not necessarily as effective on underwater images. Moreover, typical image quality metrics are not necessarily indicative of the efficacy of IE/IR when these techniques are applied prior to object recognition. In contrast, this article presents an experiment-based, quantitative evaluation of a wide range of image quality metrics and IE/IR algorithms for the purpose of underwater object recognition using laser images by comparing their resulting deep-learning-based classification accuracies. A diverse dataset of underwater laser images is acquired for seven different objects positioned at various depths and in water of five different turbidity levels. Our findings thus provide a critical review of current techniques, identify effective methods and metrics, and shed light on ongoing challenges.