Image smog restoration using oblique gradient profile prior and energy minimization

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Digital image desmogging is crucial for various imaging systems.
• Existing methods often involve computationally intensive energy function minimization and can distort image textures, edges, and colors.
• Accurate estimation of transmission map and atmospheric veil is essential for physical model-based desmogging.

Purpose of the Study:

• To develop an efficient and accurate method for removing smog from digital images.
• To overcome the limitations of existing desmogging techniques, such as computational complexity and image distortion.
• To leverage deep learning for improved estimation of physical attributes in smoggy images.

Main Methods:

• A convolutional neural network (CNN) was trained to estimate smog gradients from a dataset of smoggy and sunny images.
• Oblique Gradient Channel Prior (OGCP) was utilized in conjunction with the estimated smog gradients for image restoration.
• The CNN model focuses on estimating physical attributes necessary for effective desmogging.

Main Results:

• The proposed CNN-OGCP model demonstrated superior performance in desmogging compared to existing methods.
• The model effectively estimates physical attributes of smoggy images, leading to improved restoration.
• Performance analyses confirmed the model's effectiveness across various metrics.

Conclusions:

• The CNN-OGCP approach provides an efficient and effective solution for digital image desmogging.
• This method addresses the shortcomings of traditional models by avoiding complex energy minimization and preserving image details.
• The study highlights the potential of deep learning, specifically CNNs, in advancing image restoration tasks.