Deep Learning-Based Semantic Focus Fusion for High-Quality Multifocus Image

Abstract

Multifocus Image Fusion (MIF) is a technique in image processing that enhances image clarity by integrating multiple images taken at different focal distances into a single, sharp image. Several traditional fusion methods, including spatial and transform domain techniques, are available; however, these approaches often struggle with preserving fine details and preventing artifacts. This paper proposes a CNN-based Semantic Focus Fusion (SFF) method, leveraging deep learning architectures to improve focus region classification and eliminate blurring effects. The model is trained using a structured dataset and optimized through hyperparameter tuning to ensure efficient convergence. The effectiveness of the proposed method is evaluated using the Structural Similarity Index (SSIM), demonstrating superior fusion quality compared to conventional methods. Experimental results confirm that the proposed approach achieves high perceptual clarity, better structural consistency, and improved edge preservation. This study highlights the advantages of deep learning in MIF, offering a robust solution for applications in medical imaging, surveillance, and industrial inspection.