Recurrent Exposure Generation for Low-Light Face Detection
Mrs. M. Ratna Deepthi, M. Tech1
2Y, Sai Sowjanya, 3M. Jahnavi, 4V. Vishnu Vasundhara, 5Y. Uday Bhaskar, 6M. Joseph
1Assistant professor, Dept. of CSE, Dhanekula Institute of Engineering and Technology, Ganguru
2Student, Dept. of CSE, Dhanekula Institute of Engineering and Technology, Ganguru
3Student, Dept. of CSE, Dhanekula Institute of Engineering and Technology, Ganguru
4Student, Dept. of CSE, Dhanekula Institute of Engineering and Technology, Ganguru
5Student, Dept. of CSE, Dhanekula Institute of Engineering and Technology, Ganguru
6Student, Dept. of CSE, Dhanekula Institute of Engineering and Technology, Ganguru
Abstract— This Project introduces an innovative approach to low-light face detection, Exploiting Long Short-Term Memory (LSTM) networks, Convolutional Neural Networks (CNNs), and the Zero-Reference Deep Curve Estimation (Zero-DCE) algorithm. The primary goal is to improve the accuracy and reliability of face detection in challenging low-light conditions. Our method integrates LSTM and CNN networks with real-time exposure control, enabling adaptation to dynamic lighting conditions by capturing multiple frames iteratively with varying exposure levels. The incorporation of Zero-DCE facilitates the enhancement of exposure settings, resulting in improved face visibility and noise reduction. Experimental evaluations validate the efficiency of our approach, demonstrating significant advancements in low-light face detection accuracy compared to traditional methods. This project offers a practical adaptable solution with wide-ranging implications for real-world applications, including surveillance, security, and various other domains.
Keywords – Low-light face detection, Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNNs), Zero-Reference Deep Curve Estimation (Zero-DCE), recurrent exposure generation, real-time exposure control, dynamic lighting conditions, noise reduction, surveillance.
