Fast Single Fringe-Pattern Processing with Graphics Processing Unit
Haixia Wang,Hongxiang Zeng,Peng Chen,Ronghua Liang,Li Jiang
DOI: https://doi.org/10.1364/ao.58.006854
IF: 1.9
2019-01-01
Applied Optics
Abstract:Optical interferometric techniques provide noncontact, full-field, and high-precision measurements that are very attractive in various research and application fields. Single fringe-pattern processing (SFPP) is often required when measuring fast phenomena, which contain multiple steps including noise removal, phase demodulation, and unwrapping. However, several difficulties are encountered during SFPP, among which the processing time is of interest due to the increasing computational load brought by the large amount and high-resolution fringe patterns in recent years. In this paper, we propose a general and complete graphics processing unit (GPU)-based SFPP framework to perform a systematic discussion on SFPP acceleration. Typical methods from the spatial domain, the transform-based, and the path-related are chosen to have a variety of methods in the framework for better parallelization demonstration, namely, coherence-enhancing diffusion for denoising, spiral phase quadrature transform for demodulation, and quality-guided phase unwrapping. To the best of our knowledge, this is the first time a complete GPU-based framework has been proposed for SFPP. The advantages of performing the analysis and parallelization in framework level are demonstrated, where processing redundancy can be identified and reduced. The proposed framework can be used as an example to demonstrate the GPU-based parallelization in SFPP. Methods in the framework can be replaced but the framework level analysis, the parallel design, and the involved functions are always good references. Experiments are performed on simulated and experimental fringe patterns to demonstrate the effectiveness of the proposed work and achieve at most 29.8 times speedup compared with CPU-based sequential processing.