Owing to refraction, absorption, and scattering of light by suspended particles in water, raw underwater images have low contrast, blurred details, and color distortion. These characteristics can significantly interfere with visual tasks, such as segmentation and tracking. This paper proposes an underwater image enhancement solution through a deep residual framework. First, the cycle-consistent adversarial networks (CycleGAN) is employed to generate synthetic underwater images as training data for convolution neural network models. Second, the very-deep super-resolution reconstruction model (VDSR) is introduced to underwater resolution applications; with it, the Underwater Resnet model is proposed, which is a residual learning model for underwater image enhancement tasks. Furthermore, the loss function and training mode are improved. A multi-term loss function is formed with mean squared error loss and a proposed edge difference loss. An asynchronous training mode is also proposed to improve the performance of the multi-term loss function. Finally, the impact of batch normalization is discussed. According to the underwater image enhancement experiments and a comparative analysis, the color correction and detail enhancement performance of the proposed methods are superior to that of previous deep learning models and traditional methods.
Underwater image enhancement has drawn much attention due to the significance of underwater vision. Although considerable progress has been made in this field, a key problem remains unsolved: how can we extract and enhance minutiae while trying to remove the noise caused by scattering and attenuation? To address this limitation, we propose a new underwater image enhancement technique with a novel spiral generative adversarial framework, named Spiral-GAN, which can effectively recover real-world underwater images with more details, vivid colors and better contrast. For steady training and color correction, we include the pixel-wise losses that consist of a mean squared error and an angle error in our objective function. In addition, we design our generator with several deconv-conv blocks to preserve the details from the original distorted images. Furthermore, we present a spiral learning strategy for generalizing the enhancing model to effectively recover the real-world underwater images. Finally, we perform a number of qualitative and quantitative evaluations that suggest that our proposed approach can efficiently enhance the quality of underwater images, which can be further used for underwater object detection.
1 February 2022 in IEEE Transactions on Intelligent Transportation Systems
The vision system is important for almost all kinds of autonomous driving systems. However, visual data interfered by scattering media, such as smoke, haze, water, and other non-uniform media will be degraded seriously, showing the characteristics of detail loss, poor contrast, low visibility, or color distortion. These characteristics can significantly interfere with the reliability of autonomous driving systems. In real environments the image degradation mechanism is complex, and the estimation of degradation parameters is difficult. This issue remains to be solved. In this study, we employed dense blocks as the framework and introduced the attention mechanism to our model from four dimensions: Multi-scale Attention, Channel Attention, Structure Attention, and ROI (region of interest) Attention. With the help of the training data provided by the weakly supervised model, the proposed method achieved excellent performance in the task of scattering medium imaging optimization in different scenes. Comparative experiments show that the proposed method is robust, and is superior to other state-of-the-art methods in image dehazing, and underwater image enhancement tasks. It is of great significance to improve the reliability of autonomous driving systems in underwater and severe weather environments.
Most recent existing image generation methods have made great progress in creating high-quality images, mainly focusing on improving the generator or discriminator of convolutional generative adversarial networks (GANs). In this paper, we propose up and down residual blocks for convolutional GANs, dubbed upResBlock and downResBlock respectively. This structure is based on deconvolutions, strided convolutions, and residual blocks. With the upResBlock module for the generator of convolutional GANs, our method can further enhance the generative power of the feature extraction while synthesizing image details for the specified size. With the downResBlock module for discriminator combined with upResBlock for generator, the proposed method can speed up the back propagation of gradient and doesn't suffer from the vanishing or exploding gradients problems, generating more realistic images as well. Extensive experiments demonstrate that the proposed up and down residual blocks can help convolutional GANs in generating photo-realistic images. In addition, our method shows its universality for the improvement of existing methods.
4 May 2022 in International Journal of Machine Learning and Cybernetics