A COMPARATIVE STUDY OF CNN-BASED AND TRADITIONAL MULTI-FOCUS IMAGE FUSION TECHNIQUES WITH THE INTRODUCTION OF A NOVEL HYBRID MODEL

Abstract

This paper provides an extensive review of multi-focus image fusion techniques, highlighting recent advancements in transforming domain methods, deep learning approaches and hybrid strategies. The proposed framework combines classical multiscale decompositions such as Discrete Wavelet Transform (DWT) and Laplacian Pyramid with convolutional neural networks (CNNs) for enhanced refinement, aiming to capture intricate structural and perceptual details for superior fusion quality. Experimental evaluations reveal that hybrid models consistently outperform traditional methods in both visual fidelity and objective metrics. Despite these improvements, challenges remain, including real-time processing limitations, robustness under diverse imaging conditions and the absence of standardized benchmarks. Future research should focus on optimizing computational efficiency and developing adaptive fusion frameworks capable of addressing varied real-world scenarios.