25 September 2023 in IEEE Journal of Oceanic Engineering
Generative adversarial network (GAN)-based underwater image enhancement methods improve image quality by encoding a raw image and an adversarial image with the corresponding high-quality version. Although these methods have recently attracted significant attention, the lack of referred clear underwater imagery forces GAN-based underwater image enhancement models to be trained with synthetic or enhanced underwater images, limiting their applicability and performance. This article proposes a novel self-adversarial GAN (SA-GAN) to enhance underwater images by referring to paired raw and high-quality natural images. Specifically, a self-adversarial mode is designed that attaches a further constraint to the generation procedure. By applying two pairwise image quality discriminators, the generators are supervised with a stronger decision boundary to generate better quality than the high-quality natural image and the last-generated image. This is a novel settlement of the limitation caused by the adversary system with the synthetic or enhanced underwater images, realizing the quality transfer from natural images to distorted underwater images. Several experiments on real underwater images and two commonly used underwater image data sets demonstrate that the proposed method subjectively and objectively performs better than current methods in restoring the coloration of underwater images.
28 August 2023 in IEEE Transactions on Instrumentation and Measurement
Ranking underwater images according to their quality is a key indicator comparing the performance of different methodologies and therefore is critical in the field of instrumentation and measurement. Perceiving differences in the quality of underwater images is a challenging task that has received relatively less attention, primarily due to mixed distortion, diversified image content, and the absence of high-quality reference images. Thus, based on the pairwise underwater image quality voting data, we develop a novel ternary classification transformer to predict a quality comparison of underwater images without reference. This is the first attempt to model the quality discrimination of an image pair. The proposed model combines the perception of convolutional neural networks and Transformer encoder to explore local quality features and visual perceptual connections between different patch tokens. Experimental results reveal that the proposed pairwise underwater image quality comparison (PUIQC) scheme predicts noticeable quality differences correlating well with subjective perception. The quantification of complex distortions in underwater images compared to other learning-based methods is a compelling feature of this technology. It delivers competitive results in ranking the different enhancement outputs. In addition, we reveal the self-attention of local quality features within the two images and capture their responsive contribution to the quality decision, which explains the underlying subjective quality-sensitive mechanism during image quality comparison.