Deep Focus: Enhancing Depth of Field through Focus Stacking with U-Net Model

Limited depth of field (DOF) stands as a pervasive challenge spanning diverse fields including photography, microscopy, and computer vision, impeding accurate visual interpretation by causing blurring or out-of-focus regions within captured scenes. This paper introduces a novel solution through the Focus Stacking U-Net model, aiming to extend the depth of field in images. The issue of limited DOF is multifaceted, manifesting in macrophotography by obscuring intricate subject details and hindering comprehensive visualization, while in microscopy, it obstructs the observation of critical biological structures. Additionally, limited DOF disrupts object recognition and tracking in autonomous robotics and computer vision applications, leading to navigation errors and compromised object detection accuracy. This necessitates a concerted effort to mitigate the detrimental effects of limited DOF, thereby advancing scientific and technological endeavors. The proposed Focus Stacking U-Net model addresses this challenge by iteratively combining pairs of images captured at different focus distances, leveraging the renowned effectiveness of the U-Net architecture in image segmentation tasks. Through this iterative process, the model progressively extends the depth of field, rendering the entire scene in sharp focus. To train the model, a meticulously curated dataset comprising focus-stacked image segments is utilized, ensuring robustness and generalization capability across diverse scenes and subjects. The model demonstrates significant enhancement in DOF, providing a practical solution to challenges posed by limited DOF in imaging systems by providing a faster approach. The Focus Stacking U-Net model offers a practical and effective solution to the challenges posed by limited depth of field in imaging systems, promising to enhance visual perception and enable more accurate analysis.