“Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks”の学習報告

The quality of visual information of underwater images is mainly limited by selective absorption and scattering in water. Usually, most researches of underwater image recovery focus on addressing the scattering via inverting an optical model, but less on compensating for the unbalance attenuation caused by the selective absorption. For this reason, the recovery results are still subject to artifacts and color shifting. In this letter, a parameter-adaptive compensation method is proposed to mitigate the selective absorption. Different from the fixed parameter utilized compensation operations, the proposed method utilizes an adaptation parameter to compensate the attenuation. Further, a rigorous gradient relation between the dominant color channel and the compensated channel is established to avoid over-compensation or under-compensation. The final recovery results show the parameter-adaptive compensation method is beneficial to improve the visual quality and the Underwater Image Quality Measurement (UIQM) scores of recovered underwater images.

Parameter-Adaptive Compensation (PAC) for Processing Underwater Selective Absorption

Abstract Visual systems are receiving increasing attention in underwater applications. While the photogrammetric and computer vision literature so far has largely targeted shallow water applications, recently also deep sea mapping research has come into focus. The majority of the seafloor, and of Earth’s surface, is located in the deep ocean below 200 m depth, and is still largely uncharted. Here, on top of general image quality degradation caused by water absorption and scattering, additional artificial illumination of the survey areas is mandatory that otherwise reside in permanent darkness as no sunlight reaches so deep. This creates unintended non-uniform lighting patterns in the images and non-isotropic scattering effects close to the camera. If not compensated properly, such effects dominate seafloor mosaics and can obscure the actual seafloor structures. Moreover, cameras must be protected from the high water pressure, e.g. by housings with thick glass ports, which can lead to refractive distortions in images. Additionally, no satellite navigation is available to support localization. All these issues render deep sea visual mapping a challenging task and most of the developed methods and strategies cannot be directly transferred to the seafloor in several kilometers depth. In this survey we provide a state of the art review of deep ocean mapping, starting from existing systems and challenges, discussing shallow and deep water models and corresponding solutions. Finally, we identify open issues for future lines of research. 

/pdf/optical-imaging-and-image-restoration-techniques-for-deep-2n2hwhoe.pdf

Optical Imaging and Image Restoration Techniques for Deep Ocean Mapping: A Comprehensive Survey

An efficient learning-based method for underwater image enhancement

The topic of underwater (UW) image colour correction and restoration has gained significant scientific interest in the last couple of decades. There are a vast number of disciplines, from marine biology to archaeology, that can and need to utilise the true information of the UW environment. Based on that, a significant number of scientists have contributed to the topic of UW image colour correction and restoration. In this paper, we try to make an unbiased and extensive review of some of the most significant contributions from the last 15 years. After considering the optical properties of water, as well as light propagation and haze that is caused by it, the focus is on the different methods that exist in the literature. The criteria for which most of them were designed, as well as the quality evaluation used to measure their effectiveness, are underlined.

An Extensive Literature Review on Underwater Image Colour Correction.

The appearance of underwater scenes is highly governed by the optical properties of the water (attenuation and scattering). However, most research effort in physics-based underwater image reconstruction methods is placed on devising image priors for estimating scene transmission, and less on estimating the optical properties. This limits the quality of the results. This work focuses on robust estimation of the water properties. First, as opposed to previous methods that used fixed values for attenuation, we estimate it from the color distribution in the image. Second, we estimate the veiling-light color from objects in the scene, contrary to looking at background pixels. We conduct an extensive qualitative and quantitative evaluation of our method vs. most recent methods on several datasets. As our estimation is more robust our method provides superior results including on challenging scenes.

Unveiling Optical Properties in Underwater Images

Deep learning methods for image dehazing achieve impressive results. Yet, the task of collecting ground truth hazy/dehazed image pairs to train the network is cumbersome. We propose to use Non-Local Image Dehazing (NLD), an existing physics based technique, to provide the dehazed image required to training a network. Upon close inspection, we find that NLD suffers from several shortcomings and propose novel extensions to improve it. The new method, termed NLD++, consists of 1) denoising the input image as pre-processing step to avoid noise amplification, 2) introducing a constrained optimization that respects physical constraints. NLD++ produces superior results to NLD at the expense of increased computational cost. To offset that, we propose NLDNet++, a fully convolutional network that is trained on pairs of hazy images and images dehazed by NLD++. This eliminates the need of existing deep learning methods that require hazy/dehazed image pairs that are difficult to obtain. We evaluate the performance of NLDNet++ on standard data sets and find it to compare favorably with existing methods.

Yael Bekerman

Papers

Unveiling Optical Properties in Underwater Images

NLDNet++: A Physics Based Single Image Dehazing Network