An Efficient Attentional Image Dehazing Deep Network Using Two Color Space (ADMC²-net).

Area of Science:

• Computer Vision
• Image Processing

Background:

• Image dehazing is vital for outdoor computer vision applications.
• Existing dehazing methods often struggle with color and detail preservation.

Purpose of the Study:

• To develop a novel attention-based dehazing model (ADMC2-net) that preserves color properties and fine details.
• To improve upon the limitations of current image dehazing techniques.

Main Methods:

• The proposed ADMC2-net model utilizes two parallel densely connected sub-models operating in RGB and HSV color spaces.
• An efficient attention module with pixel-attention and channel-attention mechanisms is incorporated to extract haze-relevant features.

Main Results:

• ADMC2-net demonstrates superior performance in haze removal compared to state-of-the-art methods.
• Experimental analyses validate the model's effectiveness on both synthetic and real-world datasets.

Conclusions:

• The novel ADMC2-net model successfully addresses the color and detail preservation challenges in image dehazing.
• The proposed attention mechanisms and dual color space approach contribute to enhanced dehazing performance.