HEOPC-VGGNet Based Image Enhancement Algorithm
Chun-Hsiang Chang, Wei-Zhe Yan, Yu‐Cheng Fan · 2022
With the advancement of technology, smart cars are becoming more and more important. Smart cars combine digital cameras, Light Detection and Ranging (LiDAR), and various sensors to make their functions more and more powerful. Since the information provided by the LiDAR is limited, color images are needed to obtain more information. Nowadays, in addition to the point cloud map, self-driving technology also needs to be combined with color images to determine the correct route. Especially the color images are often too dark at night, which leads to deviations in judging routes and obstacles. Therefore, image enhancement technology becomes more important to protect the safety of self-drivers and pedestrians through nighttime image calibration. The proposed method uses the histogram equalization combined with neural networks to enhance images specifically for nighttime environments in this paper. The histogram equalization method is quite simple and fast for image enhancement, and deep learning technology has the advantage of high accuracy. In this paper, the architecture combines the characteristics of both, and is divided into two main parts. In the first part, the scheme uses histogram equalization to enhance the image, and in the second part, the enhancement factor of the image is adjusted by neural network to finally achieve a brightness enhancement, suppress overexposure, and noise in the image.