Question about interrogation window and PIV processing

[Danilo Almeida Machado]

Dear members of the OpenPIV group,

I would greatly appreciate your assistance with a question I have regarding the interrogation window and overlap in PIV processing. Specifically, I am curious about the number of cross-correlation passes that occur when using a window size of 64 pixels x 64 pixels and an overlap of 8 pixels x 8 pixels. Do the values of the interrogation window and overlap indicate that only one pass of cross-correlation occurs, or are multiple passes required due to the smaller overlap size relative to the interrogation window?

Regards,

Danilo A. Machado

[Alex Liberzon]

if you use the extended search piv function, you get only a single pass, with specified window size and overlap. default unspecified overlap is 50% of the window size. 

only windef function provides multi-pass option, but also there the window sizes should be specified, and better overlaps, for each pass. default ovelap per pass, if not specified is again 50%

great openpiv day to everyone

Alex 

---

From: openpi...@googlegroups.com <openpi...@googlegroups.com> on behalf of Danilo Almeida Machado <dmach...@gmail.com>
Sent: Thursday, May 11, 2023 6:17:43 AM
To: openpiv-users <openpi...@googlegroups.com>
Subject: [openpiv-users] Question about interrogation window and PIV processing

 

--
You received this message because you are subscribed to the Google Groups "openpiv-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openpiv-user...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/openpiv-users/CAEXGaCePQ3MzzUHo1ej2CjLbT8h7iW-EN3mKNu5gnvR6qR51rw%40mail.gmail.com.

[Danilo Almeida Machado]

Dear Professor Alex,

Hope this message finds you well. I've been using the latest version of MATLAB's OpenPIV for my research, but I'm not sure what specific function it employs.

My area of focus is velocimetry, primarily using Schlieren images. Through my observations, I have found that employing larger interrogation windows and smaller overlaps tends to produce more accurate results.

However, a question arose from my experimentation: when I use a 64x64 interrogation window and an 8x8 overlay, does the algorithm in the MATLAB version run a single pass or multiple passes?

great openpiv day to everyone

[Alex Liberzon]

In Matlab we have not implemented multipass at all. And probably will not do it. We invest all our time in Python and one team is working on C++ and merge between the two - to have a fast and accurate and open and accessible versions that do the same things with the same API.

In Matlab or Python or C++ you can use 64 x 64 with 8 overlaps - and this can be a single pass. It can also in Python be one of the steps in the multiplass (or the last step), e.g. you can do 64 x 64 with overlaps of 32, then 64 x 64 with 16 and then again with 8. I'm not sure it'll lead to any improvement. Of course you could do 128 x 128 with 32, then 64 x 64 with 8 overlaps, if you think this helps.

Alex

[Roi Gurka @ Gmail]

Hi

It seems to me there is some confusion between overlap and multi-pass: these are not necessarily related. Overlap simply uses the same information (particles) multiple times depending on the overlap sampling and mutli-pass uses prior information of a large window to allow higher resolution convergence of the velocity. If you use 64X64 with 8X8 over lap; this means you have 87.5% overlap over a each interrogation window. You essentially correlate the same particles about 8 times; it does not improve resolution but it will allow better filtering and smoothing and it should not effect the accuracy. If you use multi-pass from 64X64 to 8X8, this means you start form a large correlation window (64) and use the calculated velocity as a first guess to the next smaller window within it; for 64X64, the next step is reduction to 32X32 which generates 4 vectors from 1 vector that was extracted based on the 64X64. Than proceed to 16X16 followed by 8X8. Thus, for one 64X64 interrogation window, by using multi-pass down to 8X8, there wll be 64 vectors but with a relatively large error because there is some optimization along the path. hope this helps, Roi

[Danilo Almeida Machado]

Thank you Prof. Alex and Prof. Gurka for the assistance with overlap and multi-pass. When I experimented with different overlaps, I noticed that a lower value results in better smoothing of the spray edges. I have attached a figure to illustrate this comparison. From my understanding, applying the multi-pass function does not necessarily guarantee improved results. Selecting an appropriate overlap for the flow characteristics might be more effective in optimizing the PIV results.

Regards,

Danilo 

To view this discussion on the web, visit https://groups.google.com/d/msgid/openpiv-users/309d1c77-6071-4ba4-b598-f47204b680c5n%40googlegroups.com.