Linli Xu

This paper centers on edge-cloud-based road image enhancement under low-light nighttime conditions, with a detailed analysis of the frequency-domain characteristics of nighttime road surface images. Currently, deep learning-based night vision image enhancement methods are confronted with the extreme environment of data storms. To address this, this paper adopts an image enhancement scheme oriented toward edge node filtering. To begin with, the original image is transformed from the spatial domain to the frequency domain via the Fourier transform. Drawing on the frequency-domain attributes specific to nighttime road images, a frequency-domain perceptual filter is tailored and its filtering parameters are dynamically adjusted this enables the selective enhancement of low-frequency background contours and high-frequency fine details, while also achieving moderate noise suppression. Subsequently, the inverse Fourier transform is applied to convert the processed image back from the frequency domain to the spatial domain, yielding an initially enhanced nighttime road surface image. Following this, histogram matching is employed to further boost the contrast of the nighttime image. Finally, color balance correction is implemented to optimize the color rendering of the filtered image. Experimental results show that the frequency-domain perceptual filtering algorithm based on Fourier transform adopted in this paper cannot only effectively improve image contrast and highlight road surface texture and color information, but also meet the lightweight deployment requirements of edge nodes and can be well adapted to the hardware constraints of edge system environments.

Exploiting complementary advantages of different reflection and scattering properties of near-infrared (NIR) images and visible (VIS) images, this paper first proposes a defogging model for single image input and then develops an extended model, a fusion model for NIR and VIS color images, to enhance the visibility of image objects in outdoor atmospheric scattering environments. Our fusion model enhances the extracted details of NIR and VIS images by filtering with our defogging model in the frequency domain (F-D) that takes into account the energy preservation of these two types of images in addition to the high resolution of the fused results. Finally, based on the initial fusion, we propose a color retention mapping method to keep the fusion results free of color distortion. Experimental results demonstrate that our proposed method not only achieves good defogging effect, but also can effectively combine the complementary NIR and VIS information in image color and visibility.

Imaging research on complex samples going through random scattering medias is a very difficult challenge in the field of computational optics. With the deep learning methods applying to the field of speckle correlation image reconstruction, many problems of imaging through strong scattering media and small perturbations of the media which reduces the imaging performance have been resolved. However, due to the randomness of the scattering medium, large and complex samples are required to be trained, deep learning based methods alway occupy a lot of energy consumption on Graphic Processing Unit (GPU). Cyber-Physical-Social Systems (CPSS) integrating the cyber, physical, and social worlds is a key technology to provide proactive and personalized services for humans. This paper proposes an efficient CNN to do speckle image reconstruction based on cloud-edge computing for CPSS, which can achieve higher image resolution by less inputs.In this work, we design a self-back stacked Efficient Residual Factorized Network (SBS-ERFNet) to do image reconstruction through scattering medium. Different from the traditional ERFNets, our framework includes two stages of training, we use this model to study the speckle image from a low-to-high resolution manner. The experiments show that even if a small samples is input for training, the test results can reach a high resolution.