ADE-CycleGAN: A Detail Enhanced Image Dehazing CycleGAN Network

Abstract

The preservation of image details in the defogging process is still one key challenge in the field of deep learning. The network uses the generation of confrontation loss and cyclic consistency loss to ensure that the generated defog image is similar to the original image, but it cannot retain the details of the image. To this end, we propose a detail enhanced image CycleGAN to retain the detail information during the process of defogging. Firstly, the algorithm uses the CycleGAN network as the basic framework and combines the U-Net network’s idea with this framework to extract visual information features in different spaces of the image in multiple parallel branches, and it introduces Dep residual blocks to learn deeper feature information. Secondly, a multi-head attention mechanism is introduced in the generator to strengthen the expressive ability of features and balance the deviation produced by the same attention mechanism. Finally, experiments are carried out on the public data set D-Hazy. Compared with the CycleGAN network, the network structure of this paper improves the SSIM and PSNR of the image dehazing effect by 12.2% and 8.1% compared with the network and can retain image dehazing details.

An Efficient Attentional Image Dehazing Deep Network Using Two Color Space (ADMC2-net)

“…In this section, we perform an experimental analysis by comparing our proposed model to other existing approaches, both traditional and learning-based, TA-3DP [ 36 ], MB-TF [ 37 ], GRIDdehaze-Net [ 26 ], CMTnet [ 15 ], GEN-ADV [ 38 ], DP-IPN [ 25 ], ADE-CGAN [ 39 ]. First, we conduct a qualitative comparison to assess the compared methods on both synthetic and real-world hazy images.…”

“…In this part, we conduct a qualitative comparative analysis, by presenting a range of dehazing results for the above methods [ 15 , 25 , 26 , 36 , 37 , 38 , 39 ] including ours. In Figure 7 , Figure 8 and Figure 9 , we illustrate outcomes obtained from both synthetic (indoor, outdoor settings) and real-world outdoor hazy images.…”

“…For example, in Figure 7 e CMTnet [ 15 ] results are substantially obscure in most cases (trees in the fourth image). As depicted from Figure 7 f–h the results of GEN-ADV [ 38 ], DP-IPN [ 25 ], ADE-CGAN [ 39 ] have higher illumination, and some color shifts. Overall, as shown through all results in Figure 7 , unlike our proposed model that can recover the real scene while preserving colours and saturation, most state-of-the-art compared methods fail to restore the hazy images and are unable to preserve colours and saturation.…”

Adaptive Lightweight License Plate Image Recovery Using Deep Learning Based on Generative Adversarial Network

“…It has a value between 0 to 1. The generalised SCC formula [37] is shown in (6). This image quality assessment techniques rely on quantifying errors between reference and sample image and needs two images to do a calculation.…”

“…It consists of two parts, a contracting path, and an expansive path. There are many GAN [3] variations such as Deep Convolutional GAN (DCGAN) [4], Conditional GAN (CGAN) [5], and CycleGAN [6]. The DCGAN is usually used to generate a new image from random input data (normal distribution).…”