Shadow Detection in Single RGB Images Using a Context Preserver Convolutional Neural Network Trained by Multiple...

Area of Science:

• Computer Vision
• Image Processing
• Deep Learning

Background:

• Shadow detection is crucial for computer vision tasks like object detection and tracking.
• Existing shadow detection methods lack the required accuracy.
• Pixel-level shadow identification in single RGB images remains a challenge.

Purpose of the Study:

• To propose a novel deep learning method for accurate pixel-level shadow detection.
• To improve the accuracy of shadow identification in single RGB images.
• To develop a CNN-based model with a unique architecture for shadow detection.

Main Methods:

• A Convolutional Neural Network (CNN)-based approach for pixel-level shadow detection.
• A novel architecture utilizing a new mapping scheme in skip connections to identify global and local shadow attributes.
• Gradual utilization of multi-layer shadow context to generate precise shadow masks.

Main Results:

• The proposed method achieves superior performance on three public datasets (SBU, STC, UCF).
• Outperforms state-of-the-art methods by 3%, 6.2%, and 11.4% in Balanced Error Rates (BER).
• Demonstrates effective extraction and preservation of shadow context across multiple layers.

Conclusions:

• The developed deep learning method offers a significant advancement in shadow detection accuracy.
• The CNN architecture effectively integrates multi-level features for precise shadow mask generation.
• The training process is straightforward and adaptable for other image segmentation tasks.