Single Image Haze Removal via Multiple Variational Constraints for Vision Sensor Enhancement.

Area of Science:

• Computer Vision
• Image Processing
• Remote Sensing

Background:

• Outdoor images captured by vision sensors frequently exhibit haze-induced degradation, compromising details, color fidelity, and overall visibility.
• This degradation significantly hinders the performance of essential sensing and perception systems.

Purpose of the Study:

• To develop an effective haze-removal algorithm for enhancing degraded outdoor images.
• To improve the performance of computer vision systems by restoring image clarity and color vibrancy.

Main Methods:

• A mixed variational framework based on the atmospheric scattering model was developed, incorporating multiple regularization terms for transmission map and scene radiance.
• The algorithm utilizes ℓp and ℓ2 norms for transmission map smoothing and structure preservation, and a weighted ℓ1 norm for noise suppression in scene radiance.
• A re-weighted least square strategy was employed to solve the mixed variational model, followed by gamma correction for brightness adjustment.

Main Results:

• The proposed algorithm successfully removes haze, yielding visually satisfactory results with enhanced clarity and vibrant colors.
• Demonstrated superior generalization capabilities across diverse degradation scenarios, including low-light and remote sensing images.
• Significantly improved the performance of high-level vision tasks, such as object detection and scene understanding.

Conclusions:

• The developed haze-removal algorithm effectively restores degraded outdoor images, offering improved visual quality and robustness.
• The method shows promise for enhancing the reliability and performance of vision systems in challenging environmental conditions.