Under-Display Camera Image Enhancement via Cascaded Curve Estimation

Area of Science:

• Computer Vision
• Image Processing
• Deep Learning

Background:

• The rise of full-screen devices necessitates placing cameras behind displays, creating under-display cameras (UDCs).
• Restoring degraded images captured by UDCs presents a significant challenge in single image enhancement.

Purpose of the Study:

• To develop an adaptive image enhancement technique for UDC-captured images.
• To accurately model the complex relationship between images from under-display and display-free cameras.

Main Methods:

• A novel curve estimation network operating on Hue (H) and Saturation (S) color channels.
• Cascaded curve estimation networks with shared weights and spatial-channel attention modules for feature extraction.
• A semi-supervised learning approach combining supervised and unsupervised branches to reduce reliance on labeled data.

Main Results:

• The proposed method effectively enhances degraded UDC images, outperforming existing state-of-the-art techniques.
• Demonstrated favorable accuracy and speed, particularly for ultra-high-definition (UHD) images.
• The semi-supervised approach improved generalization to diverse, real-world scenes.

Conclusions:

• The developed curve estimation network offers a robust solution for UDC image restoration.
• Semi-supervised learning enhances model adaptability and performance with limited labeled data.
• The approach is efficient and accurate, especially for high-resolution UDC imaging.