Leveraging haze-aware features for improved image clarity and detection accuracy with an optimized DCNN-YOLOv8...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Processing

Background:

• Object detection algorithms struggle in foggy environments due to low visibility and contrast.
• Existing Image Quality Assessment (IQA) algorithms have limited generalization for hazy outdoor images.

Purpose of the Study:

• To develop novel techniques for IQA and object detection specifically for hazy outdoor images.
• To create an integrated architecture addressing limitations in foggy environment object detection.

Main Methods:

• Proposed Haze aware Structural Pixel Neighbor (HSPN) features, Color rendition, and Mean Subtracted Contrast Normalized (MSCN) coefficients for hazy images.
• Image quality assessment using a Deep Convolutional Neural Network (DCNN) trained with the Chronological Chimp Optimization Algorithm (CChOA).
• Object identification using an enhanced YOLOv8 detector integrated with the haze-aware image quality assessment.

Main Results:

• The integrated system demonstrated enhanced performance in object identification within foggy conditions.
• Achieved high performance metrics: Mean Seismic Data Structural Similarity (MSDSS) of 0.944, Signal-to-Noise Ratio (SNR) of 50.769, and Structural Similarity Index (SSIM) of 0.925.

Conclusions:

• The developed architecture effectively improves object detection in foggy environments.
• The novel CChOA and enhanced YOLOv8 provide a robust solution for IQA and detection in challenging visibility conditions.