Year: 2024
Type: conference-paper
Abstract: In view of traditional image improvement technology's limitations in high real-time application environments, a new low-light video image improvement algorithm is proposed in this article. In consideration of traditional improvement algorithms' faults, and under the precondition of studying low-light improvement techniques in image signal processing (ISP) systems, deep learning frameworks are introduced to break through complex structure and long processing pipeline faults in traditional ISP systems. In this article, a lightweight image brightness improvement algorithm with a basis in convolutional neural network (CNN) is proposed. In this algorithm, feature extraction of RGB images is performed through CNN to generate a brightness coefficient matrix, simplifying RGB to RGB direct mapping and reducing model numbers, model parameters, and computational complexity. This technology pays attention to generating coefficients of brightness, can effectively extract changing parts of brightness, maintain detail information, and conduct efficient and natural brightness improvement, and can apply to high real-time application environments such as autonomous driving.
Authors: Siqi He, Jianping Xing, Yu Bai, Shuzhen Huang, Kaiping Sun, Peng Sun
Language: English
Topic: Image Enhancement Techniques
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4409833002