This article investigates the problem of enhancing underwater visual observations for the purpose of accurate underwater object detection. Most existing underwater visual enhancement algorithms tend to follow human vision preference but do not necessarily favor the effectiveness of an object detection algorithm. We observe that it should not be the human vision preference but the object detection algorithm that knows what underwater visual enhancement configuration is most beneficial to the detection tasks. In light of this observation, we propose a reinforcement learning paradigm of configuring visual enhancement for object detection in underwater scenes. Specifically, we use underwater image features as states and object detection score increments as rewards. We set up a collection of extensible actions that consist of multiple visual enhancement algorithms. The optimal policy is learned in the form of an action sequence, which characterizes a stepwise process of visual enhancement. Experimental results validate that the sequence of visual enhancement algorithms configured with respect to the object detection algorithm is in favor of improving the detection results.

We observe that a natural image tends to exhibit similar histograms for color channels in the RGB color space and consistent statistical estimates for color channels in the Lab color space. We refer to these observations as natural color consistencies. In contrast, we discover that an underwater image does not always follow the natural color consistencies. Different color channels in an underwater image tend to give rise to very different distributions, regardless of whether the channels are in the RGB color space or Lab color space. We refer to these observations as underwater color disparities. To enhance an underwater image to make it appear more natural, it is necessary to correct its underwater color disparities to align with the natural color consistencies. To this end, we develop an adaptive attenuated channel compensation method based on optimal channel precorrection and a salient absorption map-guided fusion method for eliminating the color deviation in the RGB color space. We then develop a method to enhance the contrast of channel L and an adaptive color distribution specification method for improving the contrast and matching the color distribution in the Lab color space. Additionally, we develop an edge-enhanced mask fusion method for correcting blurry details. Our method is not a deep learning method but can effectively be applied to a single underwater image. The qualitative and quantitative empirical results validate that our method outperforms state-of-the-art underwater image enhancement methods. We release the reproducible code at https://gitee.com/wanghaoupc/Underwater Color Disparities for public evaluation.

In this article, we develop an underwater image enhancement framework based on reinforcement learning. To do this, we model the underwater image enhancement as a Markov decision process (MDP), in which states are represented by image feature maps, actions are represented by image enhancement methods, and rewards are represented by image quality improvements. The MDP trained with reinforcement learning can characterize a sequence of enhanced results for an underwater image. At each step of the MDP, a state transitions from one to another according to an action of image enhancement selected by a deep Q network. The final enhanced image in the sequence is obtained with respect to the biggest overall image quality improvement. In this manner, our reinforcement learning framework effectively organizes a sequence of image enhancement methods in a principled manner. In contrast to the black box processing schemes of deep learning methods, our reinforcement learning framework gives a sequence of specific actions, which are transparent from the implementation perspective. Benefiting from the exploration and exploitation training fashion, our reinforcement learning framework possibly generates enhanced images that are of better quality than reference images. Experimental results validate the effectiveness of our reinforcement learning framework in underwater image enhancement. The code and detailed results are available at https://gitee.com/sunshixin_upc/underwater-image-enhancement-with-reinforcement-learning.