Generalized Parallel Implementations of Intuitive Digital Focus Indices with Cross-Platform Compatibility

• HyungTae Kim,
• Duk-Yeon Lee,
• Dongwoon Choi,
• Dong-Wook Lee,

Vol. 19, No. 11, November 30, 2025
10.3837/tiis.2025.11.015, Download Paper (Free):

• Automatic focus
• Digital Focus Index
• GPU
• multi-core processor
• Parallel Processing

Abstract

Automatic focus (AF) is indispensable for maximizing the usability of image sensors in mobile devices and industrial automation. However, the computational cost of AF has increased significantly owing to the high-resolution images and multiple image sensors used. Thus, we present generalized parallel architectures of intuitive digital focus indices with reduced complexity and parallel kernels implemented with open-sources for cross-platform compatibility. The thresholded pixel count, thresholded content, and image powers based on the sum of the powers of a threshold are explored to reduce computational costs and mathematical complexities. Generalized parallel architectures are designed to focus on the threshold, power, and sum using a graphic processing unit and a multi-core processor. The parallel kernels are evaluated using various image resolutions, pixel bit-depths, and computing platforms. Compared with conventional AF approaches that employ sequential processing, the average acceleration is frequently larger than 5.53×, and the maximum acceleration was 27.90×. The results demonstrate significant improvements in processing speed in diverse computing platforms, which implies time-efficiency and energy-savings. The cross-platform kernels can be integrated with the subroutines in real-time imaging, image denoising, signal processing, and artificial intelligence.

Statistics

Cite this article