DRPL

DRPL performs multi-focus image fusion using a deep learning pair-learning framework that processes entire images to generate binary focus masks distinguishing focused and defocused regions and produce all-in-focus fused images for image processing and bioinformatics applications.

Key Features:

• Whole-image processing: Processes entire source images directly without dividing them into small patches to determine focus status.
• Pair learning strategy: Uses pairs of complementary source images as inputs to learn corresponding focus masks for each image in the pair.
• Binary mask generation: Produces binary masks that distinguish focused from defocused regions for each input image.
• Complementary constraint: Imposes a complementary constraint on paired images to ensure fused output maintains focus coverage across regions.
• Gradient loss: Incorporates a gradient loss function that leverages edges and gradients present in focused areas while accounting for their absence in defocused regions.
• Structural Similarity Index (SSIM): Employs SSIM to balance fidelity between reference images and the final fused output.
• All-in-focus fusion: Produces fused images intended to be visually all-in-focus by combining complementary focused regions.

Scientific Applications:

• Multi-focus image fusion: Applied to fuse complementary images with varying focus into a single all-in-focus image.
• Image processing and bioinformatics: Used where all-in-focus reconstructions are required for downstream image analysis in image processing and bioinformatics contexts.
• Benchmarking on datasets: Evaluated on synthetic and real-world datasets, demonstrating effectiveness relative to existing methods.

Methodology:

Processes entire image pairs with a deep learning pair-learning approach to generate complementary binary focus masks, enforces a complementary constraint, and optimizes model parameters using a gradient loss that emphasizes edges together with SSIM to balance reference and fused outputs.