Digital Image Correlation and Tracking

Calculate displacement and strain from a series of images

Description

Compilation of .m files to calculate strain from images independent of physical size of the samples. Strain can be calculated in horizontal as well as vertical direction. The functions were developed at the Johns Hopkins University by Rob Thompson, Daniel Gianola and me, all of us part of the Group of Kevin Hemker. Dan is now at the UPenn (http://gianola.seas.upenn.edu) and I am now at the Karlsruhe Institute of Technology (http://www.yin.kit.edu/?id=57, Germany) and will continue to work on the code. Sven Bundschuh just entered the authors guild and provided a new grid_generator.m and some essential changes.
If you have special questions please check the header of the .m file and email the responsible person:

[e-mail]
[e-mail]

Have fun with the functions.

We used only windows systems so far. Linux Users need to copy cpcorr.m and findpeak.m into their work folder and change it there. After renaming it (e.g. cpcorr2.m) the calling lines should be changed, too, in the functions automate_image.m and RTCorrCode.m. Parallel Computing Toolbox is optional.

Acknowledgements

This file inspired Shear Strain Dic (For Bias Extension Tests), Digital Image Correlation 4 Threads Pmode, Optical Strain Measurement By Digital Image Analysis, and High Corr High Temperature Digital Image Correlation Software.

• Required Products: Image Processing Toolbox; Optimization Toolbox
• MATLAB release: MATLAB 7.10 (R2010a)
• Other requirements: We used only windows systems so far. Linux Users need to copy cpcorr.m and findpeak.m into their work folder and change it there.

[Jesus Emilio]

 

[Rob Pierce]

 

[Elizabeth Jones]

Hi Chris,

First, I want to thank you for putting together this set of code, documenting it, and posting it for others to use. I have been using it, and modifying it, for almost a year now, and have found it to be overall a very decent program.

I have made many changes to the set of .m files, most importantly: calculating displacements for EACH image, calculating strains using finite element shape functions, making GUIs to run the code in a more streamlined manner. I plan on documenting the changes and the new code, and hope to post them on Mathworks when I get a final version.

More recently, I have started looking into how to speed up the code. As the number of grid points in each image increases, and the number of images to correlate increases, the automate_image function can take a lot of time to run, nearly all of the time spent running cpcorr. As cpcorr was written by folks at Mathworks, it is pretty well optimized already. However, improvements can be made by parallel computing.

I have started to revise the original code using a parallel for-loop to loop over the images in the automate_image function. There is some improvement, but not as much as I’d like. I searched for ways to parallelize the cpcorr function itself, and I came across a poster on Nvidia’s website (link below) that describes optimizing your original DIC code to run on a GPU. Though the authors reference this File Exchange in the poster, I don’t see that they posted a reply to this thread. I wanted to make all users of this DIC code aware that there is a version that, when run properly on a GPU, should run faster than the original code.

Poster: http://www.nvidia.com/content/GTC/posters/2010/I19-GPU-Computing-for-Real-Time-Optical-Measurement-Techniques.pdf

Cuda optimized DIC code: http://dside.dyndns.org/dict

Cheers,
Elizabeth

[Enrico]

Hi Chris, Thanks a lot for your effort!
I have the same problems of maryam abyaneh:

please help me with these:
1- I have 10 slightly displaced pictures but I only get 8 images after DIC.Why?

2- How can I get the result dispalcement for each image?

Thank you very much!

[Chris]

Hi Tom,
you can find them here:
http://www.mathworks.de/matlabcentral/fileexchange/12456
Chris.

[Tom]

I'm sure I'm missing something really obvious, but I can't seem to find any sample images to test this code. Can anyone point me in the right direction?

[Deepesh Goel]

I tried to use the code for the sample images provided with the code. But the answers I got are different in order of magnitude when compared to answered obtained from other codes. Is it because of the different algorithm or approach applied. Please mention the name of approach applied in this. Also what is the reference image it uses? Please help me!! Thanks in advance

[shah baba]

'DIC_test' can't be seen.
Thanx for nice code. I am sure that you have alos submited your sample images, but unfortunately I can't see after unzipping the 'DIC_test' folder. Any idea?

[shah baba]

Thanx for nice code. I am sure that you have alos submited your sample images, but unfortunately I can't see after unzipping the folder. Any idea?

[RenO]

Interesting programming. But I have a question of how to clearly interpret the displacement plots. The markers shown relate to those markers from the grid created on the images, so not all markers from the grid move? Is there a post-processing guide to help with analysis?

Thanks.

[LIN]

This is a powerful code! Could you plz tell me how to add interpolation function (like cubic spine etc.) thank you.

[Rodrigo Oviedo]

I believe there's an error in the displacement.m function. On line 936, it says "maxminusminvalidy=(max(max(validx))-min(min(validy)));", while it probably should say "maxminusminvalidy=(max(max(validy))-min(min(validy)));".

[Michael Versteeg]

I have successfully ran automate_image.m with an 8bit TIF image but I would like to use the original 14bit image taken from the camera (.b16 file format)to get a more accurate strain distribution around a crack tip. The first option I tried is to save the image histogram into a 16bit TIF which made the image in Matlab completely black and the readings inaccurate. I think this was due to a large portion of the TIF file being blank. To fix this the histogram was spread over the 16bit which corrected that and made it visibly clearer then its 8bit counterpart.

While running automate_image.m with the new 16bit TIF the grid does not follow the image set past the first two images. Do you know why this could be?

[Vaidehi Jakkali]

Hi

Can this software be extended to do 3D DIC?

[Christoph Eberl]

Hi Ali,

I guess you need to check for face detection algorithms, the DICT code won't work for this purpose.

Chris.

[Electron Ali]

Sir , i m interested in Face detectioN of A particuLAr perSon, in a crowd.....

Example:.. i haVe daTa stored in my a Computer,,of a particular person or a grp...,,,.,.then my live camera detect a face frm a crowd ,who match with, a persons, in my list,,,

i get automatically its complete data and info,,.,.

i m electronic student,.plZ,.,help me,
u gUys help me tOO....
([e-mail])

SALAM

[Electron Ali]

Sir , i m interested in Face detectioN of A particuLAr perSon, in a crowd.....

Example:.. i haVe daTa stored in my a Computer,,of a particular person or a grp...,,,.,.then my live camera detect a face frm a crowd ,who match with, a persons, in my list,,,

i get automatically its complete data and info,,.,.

i m electronic student,.plZ,.,help me,
u gUys help me tOO....
([e-mail])

SALAM

[Steve Mills]

Is there an image size limit? My images are 18megapixel (5180x3450). Do I need to adjust an array size?

I also get the following error.
>> automate_image_mp_2009b
??? Error: File: automate_image_mp_2009b.m Line: 102 Column: 4
MATLAB cannot determine whether "labindex" refers to a function or variable.
See SPMD in MATLAB, "Overview".

[Christoph Eberl]

Hi Lorraine,

you can find cpcorr in the folder for the image processing toolbox (../matlab/images/images/private).

Chris.

[KYAW KYAW]

Dear Chris,

I tried to download your zip file but It won't. Pls advise me and I could get the other files last time in here.

Thanks and best regards

[Lorraine Brady]

hi
i was just wondering where the m file cpcorr can be found. Its prob somewhere very obvious as know one else seems to have the same problem but as of yet i can not find it.
Thanks
Lorraine

[kirit sharma]

 

[Chris]

Hi Asheesh,
we do reduce the displacement noise by averaging over time. That helps a lot. Although we do not use the contour plot that often but typically pick the region we want to know the strain in and process it there. I will add an .m file which can do that.

The other way would be to measure the local strain from the next neighbours, which would be easy to implement but I haven't had the time to write it down properly. If there is much need I can try and fit it into my schedule.

Cheers, Chris

[monruedee]

Hello, My name is Monruedee I am a student at GIST, South Korea.
My reacher is reverse osmosis membrane fouling investigation using AFM and SEM . I found your code and I wonder that is it possible the apply your code to tract out infomation from Fouled membrane image (bothe of SEM and AFM). Since I lack of knowledge about MATLAB. Please give me some suggestion. My email is "[e-mail]"

[Asheesh Lanba]

Hi Chris,

My congratulations on an extremely easy to use and efficient code. It has made my life much easier.

One problem that I face though is that no matter what I do, I can't seem to get smooth contours using 'Full Strain Plots'. I tried deleting bad markers but nothing seems to work. Do you have any suggestions?

Regards,

Asheesh

[sudhir sharma]

 

[Jay]

For some reason, I can't get it to give me output more accurate than 0.1 pixel. Even though I changed lines 134 and 135 of cpcorr exactly how it says, I still get 0.1 pixel precision. I tried this many times on different computers and I've checked it is correct many times. Any suggestions?

[Jay]

 

[John]

 

[John]

 

[mano]

Hi Everybody,
I posted this question before and I did not find any answer. Could anyone who knows the answer kindly write me back.
I would like to know that if any calibration in the software that has been genoursly posted by Christopher is needed or not??? Actually, I am asking this question because, I am using of this software for determination of the displacements in a beam with a notch.
Thank you,
Manoo

[Sampath Vanimisetti]

Christoph,
Thanks for the receptive response to my previous comments. To you point, I did realize the need for seperating the algorithms for small from large deformations and have made some modifications to the code. If you think my code edits and images for large strain expts would help, I can send them over.

Thanks again,
Sampath

[mano]

Hi Christoph,
Thank you for your software.
I am wondering that there is no calibration in your software or I could not find it... would you please help me about this? Because to the best of my knowledge this is a very importamant step.
Thank you

[Christoph Eberl]

Hi Sampath,

you're absolutely right and I very much like the experiment you conducted. Actually your point is always relevant if either the surface or the illumination is changing. After all everything depends on that. Rob Thompson used the floating reference image in his software as far as I remember(file id 14850) to deal with the changes in illumination when going from RT to 1200°C.
In case of large deformations your changes are going to make the code work much better. For small deformation that would introduce some error, therefore it would be useful if the code either decides by itself to apply this method or the user can decide at the beginning. We will put that into the next version of the code.
The same problem comes up in case of rotation to which DIC is extremely sensitive. Other codes are capable of taking the deformation of the image into account by measuring the local deformation and applying it to the virtual markers. That would be something for the future. We will also look into that and help is always welcome. Contributions will be cited!

[Sampath Vanimisetti]

Hello Christoph,

Firstly, I would like to thank you, Rob and Daniel for putting together this nice tool. I checked the tool out of curiosity and found it has been put together very well. Kudos!

However, if I may, I felt there is scope for some minor improvement. The current version of you code uses reference (or first) image for correlation. This may be sufficient for small strains where the initial correlation kernel (or convolution kernel in some instances) does not alter from frame to frame. However, deformations involving large strains, especially those in polymeric or rubbery materials, significantly distort the correlation kernet between the base and input image. In such cases, the control-point correlation to track displacement may drop some points off the grid. This was observed even if the kernel size was increased to 20 (instead of the recommended 15).

I tried this little experiment at home with a section cut from a used Scotch Brite (3M) kitchen glove. On the glove, I printed a square patten of random dots with a ball point pen. Acquired images using Fujifilm S500 connected to home computer in PC-Cam mode, while the piece of glove (I guess silicone rubber) was deformed to about 100% strain (did not have anything fancy at home, so used my good old hands!). Now, once I had all the images I ran it through your DIC program and found that some points are dropped off the grid for reasons stated above.

To this end, I made changes to your code (very minor), by switching from total deformation to incremental deformations. This, in relation to the algorithmic implementation would mean that control point correlation should be performed between the current frame and the previous frame in the deformation histroy, instead of the initial (first) frame. The correlation as a result was found to give better results.

If you want, I can send you the updated code along with all relevant files. I also have some additional stuff that you might want to add to your file list here.

Let me know your thoughts & thanks for the great project!

Best, Sampath.

[Christoph Eberl]

Hi TRAN,

this depends on the the size of the subset you are using. You set the size in cpcorr.m . With the initial set to corrsize=5 pixels the detection is roughly 2-3 pixels, if you turn it up to15 or 20 we were able to track jumps of 15 pixels (due to cracking of a sample). If the jumps are too large, you can reduce the size of the image. You could also alter the code to run a rough analysis on a reduced image size before you run the dense analysis. We were able to detect jumps of up to 100 pixels that way. Unfortunately the code is not as polished and prooven as it should be before we release it on matlab.

Cheers, Chris.

[TRAN]

Hi Christoph
I would like to know that how much pixels the function cpcorr is able to detect the movement of the cross at least and how to set up to reach this level there.
Thank you very much

[Christoph Eberl]

Hi Mingkai;

for 2 images you can also use the cpcelect to register two images.
Our code should work also with two images but will only save the displacement in validx and validy which you have to compare to your reference grid_x and grid_y by hand. The displacement command needs at least 3 images to show the displacement. To get around that just make copies of your images and run the code e.g. on four images. That will work fine.

Chris.

Chris.

[Christoph Eberl]

Hi Brett,

no we do not use interpolation techniques, but you could easily alter the code. Acording to literature it is better to use a linear or Gauss filter to soften the image. We do that in a extra step beforehand to have a control of what the code is doing to the raw data.

Instead we are taking the mean of the calculated marker displacement (e.g. taking 5 images per second and then take the mean of 5 consecutive images in the validx - see code segment).

[nummarker numimages]=size(validx);
numaverage=5;
for j=1:numaverage
validxtemp(:,:,j)=validx(:,j:numaverage:(numimages-numaverage+1)+j);
validytemp(:,:,j)=validy(:,j:numaverage:(numimages-numaverage+1)+j);
time_image_temp(:,:,j)=time_image(j:numaverage:(numimages-numaverage+1)+j,:);
end
validx_mean=mean(validxtemp,3);
validy_mean=mean(validytemp,3);
time_image_mean=mean(time_image_temp,3);

This shows much better results as we get also rid of vibrations.

Chris.

[Christoph Eberl]

Hi Andy,

this seems to be a problem with our code. Do you have the newest version? We run 2009a and everything seems to work. Can you post the error if this is still a problem?

Chris.

[Christoph Eberl]

Hi Deva,

our 2009a runs on Windows and does work as far as I can tell. Do you get any error message?

Chris.

[Mingkai]

I analyze two consecutive images using this code. I can not get the results at all. Do anyone face up with this problem as well?
Thanks, Mingkai

[Andy Rix]

I'm also running 2007b and having problems at the grid_generator stage as it says some of the syntax is obsolete. Is this an issue with the script or my TIF images?

Thanks, Andy

[Deva]

Has anyone used this on Mac/Unix R2009a yet? I seem to be running into some issues with automate_image.m. The figure window comes up but isn't populated with an actual figure. Any insight would be appreciated!

[Brett]

I was wondering what interpolation technique (if any) is used for points with a small SNR.

[Christoph Eberl]

Hi all,

sometimes matlab does have the old files in the cache. I think after a cache refresh and a check if the old version is still there the problem can be solved.

Cheers, Chris.

[Megan]

Oh, BTW, I'm running 2007b. Has anyone got the most recent update (11 Sept 2008) to run on 2007b yet? Thank you, Megan

[Megan]

Hi there, my question is directed for Chris and Rob since my errors appear with running automate_image. I receive the following error message:

In imuitools/private/initSize at 75
In imshow at 239
In automate_image at 80
??? Error using ==> obsoleteHMS at 7
Function TIMEDIM is obsolete.

Error in ==> timedim at 16
obsoleteHMS(mfilename,'')

Error in ==> automate_image at 135
[h,m,s] = hms2mat(timedim(estimatedtime,'seconds','hms'));

The code at line 135 of automate_image never seems to use the hms2mat function so I'm confused in my effort to alter the code so it will run successfully. Everything up to this point was extremely clear in the "Correlation_Guide_2" so I'd really appreciate it if anyone had any suggestions to help me along. (I changed the "imshow" calls to "imtool" but I don't think this should make a difference.)

Thank you!
~Megan

[liu xiao-yong]

I am a beginner.so I will try my best to study.

[Chris Eberl]

Hi Julie,
sorry for the long delay, Itry to establish myself at University in Karlsrueh, Germany, and strating a group eats up more time than expected ;-).
I will post new code tonight which will fix that incompatibility with 2007a/b.
Chris.

[Chris Eberl]

Hi Shafqat,
you still need to process validx and validy with displacement.m. RTCorrCode does nothing else than take the slope of the displacement which is stored in the named files. Please let me know if you need more information. Chris.

[prashant gudi]

nice

[Shafqat Rasool]

How we can get Strain data file from RTCorrCode.Because it generates only strain vs. Image # graph...help me please

[Julie Waters]

hi, this doesn't seem to be working in R2007b yet, but seems ok in R14. any further updates in the works?

[ali haddy]

 

[Christoph Eberl]

Hi Adam,
yes it will. I'm in the process to ex-out all these issues since I got the 2007b at last. I'm on it right now and will see if I can upload it in the next few days.
Cheers, Chris.

[Adam B]

Excellent. Will the coming update fix the problems of calling obsolete functions in the newest matlab release?

[Christoph Eberl]

Functions were updated today and a few new functions added to the package. Should be up and running in 1-2 business days. Correlation guide is not yet updated. Other updates will follow soon.

Please report any problems. Have fun, Chris.

[Anonymous guest]

A free software that works great. Thanks.

[ihsan ihsan]

Great job. Thanks for sharing.
regards

[aa mo]

very good

[Christoph Eberl]

Functions were updated today and new functions added to the package. Should be up and running in 1-2 business days. Please report any problems. Have fun, Chris

[shakiba moaddab]

 

[Christoph Eberl]

Under the title:
'Digital Image Correlation and Tracking Example Files and Slides'
you will find soon the images used in the Digital Image Correlation and Tracking Guide included in the Image Correlation Package. Furthermore I packed the Powerpoint slides into the zip file (although in low resolution) which explain the whole project and give some hints how experimental setups could look like and under which conditions correlation or tracking works best. Chris.

[John D'Errico]

Thank you for adding this extensive help. This code would be unusable without that help file.

[Christoph Eberl]

We're happy about the resonance and take the critical comments serious.
The reason why the guide was not in the package is that matlab does not accept pdfs. I updated the zip file with a html version of the guide. Should be online in a couple of hours.
I can provide the slides as jpgs in low resolution or as pdf by email (see above) if you're interested.

[Frederic Moisy]

Submissions on the File Exchange should be documented and should contain example. This submissions contains zero help and zero example. In itself, this submission is useless. I don't see any good reason why Matlab Central users should contact the author to get additional meterials that should have been included with the submission. Please update.

[Pablo Rivas Perea]

Excellent work, if you ask for the guides or the slides, you'll get enough help to understand the scope of the project. The authors gives a very useful approach on this concern.
Good for you!

[viet nguyen]

viet

[Frederic Moisy]

Seems to be interesting, but there is absolutely no help! I see that you mention "If you're interested in the code please write me an email to get a copy of a guide and some slides explaning the basics". But please submit only complete and documented files here! You should definitely include an help, a 'Contents.m' file, and some examples with sample data in your zip submission.

[Shailendra Joshi]

I have seen the utility of this suite of functions in the accurate description of two/ three dimensional strain states in a specimen under load. Chris and his team at JHU have put up a great software program that utilizes Matlab capabilities very effectively.

Congrats!!