MLEFGN

MLEFGN denoises images by using a multilevel edge features guidance network that predicts edges with an Edge-Net CNN and leverages a dual-path architecture to preserve structural information during denoising.

Key Features:

• Edge Reconstruction Network (Edge-Net): A CNN that predicts clear edges from noisy images and provides edge priors for guiding denoising.
• Dual-path Network Architecture: Simultaneously extracts image and edge features to leverage both types of information for noise removal and detail preservation.
• Multilevel Edge Features Guidance Mechanism: Utilizes edge priors generated by the Edge-Net across multiple levels to guide the denoising process and maintain structural clarity.
• Edge Detection and Guidance: Embeds the Edge-Net within the denoising architecture so edge information is preserved and utilized throughout processing.
• End-to-end Learning: Integrates edge detection and image denoising in an end-to-end trainable framework.
• Robustness and Accuracy: Demonstrated robustness and accuracy through extensive experiments, showing superiority over traditional methods.

Scientific Applications:

• Natural image denoising: Enhances denoising of natural images while preserving edge integrity and structural details.
• Medical imaging: Supports applications that require high-quality denoised images with preserved anatomical edges.
• Remote sensing: Improves denoising for remote sensing imagery where edge preservation is critical for feature interpretation.
• Computer vision: Provides improved input quality for downstream computer vision tasks that depend on accurate structural information.

Methodology:

MLEFGN embeds an Edge-Net CNN for edge detection and guidance, employs a dual-path architecture with multilevel edge feature guidance, and is trained end-to-end; extensive experiments validated its robustness and accuracy.