An infrared night vision image enhancement algorithm based on cross-level feature fusion

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Infrared night vision images suffer from low quality due to insufficient light, leading to color overflow and discontinuity.
• Single feature fusion methods often result in larger halo areas and lower Peak Signal-to-Noise Ratio (PSNR) after enhancement.
• Existing enhancement techniques struggle to effectively address noise and enhance visual details in challenging low-light conditions.

Purpose of the Study:

• To propose a novel infrared night vision image enhancement algorithm.
• To address the limitations of existing methods in handling noise and improving visual quality.
• To achieve high-quality enhancement of infrared night vision images through advanced feature fusion.

Main Methods:

• Denoising infrared images using smooth wavelet decomposition and a neighborhood-based wavelet coefficient shrinkage algorithm.
• Preliminary image enhancement via Retinex algorithm, bilateral filtering for illuminance estimation, and Sigmoid function for reflection enhancement.
• Constructing a cross-level feature fusion network for multi-level feature extraction, reconstruction, and adaptive fusion, optimized with a joint loss function.

Main Results:

• The proposed algorithm effectively reduces noise interference in infrared images.
• Preliminary enhancement improves the overall visual effect by addressing illuminance and reflection.
• The cross-level feature fusion network further enhances feature information, leading to high-quality image output.
• Experimental results demonstrate PSNR values exceeding 30dB and Structural Similarity Index Measure (SSIM) values over 0.73.

Conclusions:

• The developed algorithm significantly enhances infrared night vision images.
• The cross-level feature fusion approach proves effective in improving image quality and visual perception.
• The method achieves superior performance and high enhancement effects compared to existing techniques.