Surrogate-based cross-correlation (SBCC)

This repository contains code for the submitted paper Surrogate-based cross-correlation for particle image velocimetry. In this work, the SBCC is proposed to improve the cross-correlation performance via an optimized surrogate filter/image. Our broad SBCC encompasses several existing correlation techniques(PC, SPOF, RPC, etc) as special cases. Besides, the SBCC demonstrates the best robustness by incorporating other negative context images.

Motivation

Inspired by correlation filters, the surrogate image---supplanting original template images---will produce a more robust and more accurate correlation signal. That says, the surrogate is encouraged to produce a predefined Gaussian shape response to Image 1, and zero response to negative context images. As a result, the response between Image 2 and the surrogate could be accurate and robust. Our SBCC framework is significantly different from the existing correlation methods by considering other negative context templates. More detailed info is referred to the paper , paper link will work once available.

Install dependencies

conda install numpy matplotlib opencv seaborn

The experiments

• Exp1.ipynb: Visualize the cross-correlation response map for synthetic images;
• Exp2.ipynb: Visualize the cross correlation response map for a real PIV images;
• Exp3.ipynb: RMSE test with noise & w/o noise, uniform flows;
• Exp4.ipynb: Test on the synthetic dataset of non-uniform flows;
• Exp5.ipynb: Test on the real PIV cases;

Questions?

For any questions regarding this work, please email me at [email protected].

Acknowledgements

Parts of the code in this repository have been adapted from the following repos:

• OpenPIV/openpiv-python
• opencv/opencv