Proceedings Article | 20 November 2024
Chengliang Fu, Sen Han, Ying Yang, Yuhang Shen, Liwei Zhang, Yanjun Wang, Linji Ji, Jiaqi Qin
KEYWORDS: Phase unwrapping, Parallel computing, Mathematical optimization, Windows, Super resolution, Image processing, Design, Algorithm development, Signal processing, Laser interferometry
With the continuous development in the field of optics, high-resolution and even super-resolution image processing has become key to improving image quality and resolution. However, phase unwrapping algorithms are crucial in interferometry, but their complex computational methods often lead to inefficiencies, especially as CPU integration increases. Despite this, single-thread performance is constrained by factors such as power walls, frequency walls, and instruction-level parallelism, resulting in low efficiency. To address these limitations and enhance image processing speed, this paper proposes a high-performance CPU computing algorithm based on the Goldstein algorithm, combining OpenMP technology with SIMD instructions. In the three critical steps of Identification of Residues, Branch-cutting, and Integration, tasks are divided into multiple subtasks and executed simultaneously by independent computational units within a single clock cycle, solving problems more quickly, especially during the Integration process where a new algorithm is introduced. To evaluate the optimized algorithm, we conducted multiple tests on super-resolution images. The results show that as pixel size increases, the algorithm optimized with high-performance CPU computing demonstrates significantly better performance than standard CPU computation. On an RTX 3060 laptop, using a phase map with a resolution of 9344×7000, we achieved a 13.5-fold speed improvement. Therefore, combining this algorithm with CPU parallel computing can significantly enhance the efficiency of interferometry.
