I'm analysing some imaging data but having a bit of a problem. When the images were taken, some of them have uneven shading. I want to set a threshold to analyse the area certain features within the image but have to adjust manually for each part of the image and between images to capture the parts that I want.
I have heard people said to keep the threshold the same between images. But I also found some protocols also suggest that if there's uneven shading, manually adjustments will be needed.
Even between images that looks very similar, when I apply the same threshold, I always have to readjust it too because the threshold for one image won't match 100% with the next.
Do you think it's ok to do so? Many say this will introduce biases but I personally think the biases will not be as big and that it's still better than keeping the same threshold between images of different shading that will capture unwanted parts.
The uneven shading seems unavoidable no matter how careful I am.
Thank you very much
Rolling ball background subtraction in ImageJ does a great job of correcting uneven illumination. It is in the "Process" "Subtract Background" menu option. Set the Rolling Ball Radius to be at least as large as the radius of the largest object in the image.
You may classify the images first then a framework of thresholds could be used for different groups of images.
Rolling ball background subtraction in ImageJ does a great job of correcting uneven illumination. It is in the "Process" "Subtract Background" menu option. Set the Rolling Ball Radius to be at least as large as the radius of the largest object in the image.
Dear Anh,
This is a well known problem in image processing, and the solution is called adaptive thresholding. These are algorithms that will determine a local (relative) threshold based on smaller portions of your image. This way, you don't have to rely on one suboptimal threshold for the entire image. Since these methods are standard, and the results are reproducible, you can refer to them without having to worry about bias that manual thresholding may bring.
The following pages give an overview of how it works. There are multiple plugins in ImageJ that you can use, pick your favorite.
http://homepages.inf.ed.ac.uk/rbf/HIPR2/adpthrsh.htm
https://sites.google.com/site/qingzongtseng/adaptivethreshold
http://imagej.net/Auto_Local_Threshold
http://www.gcsca.net/IJ/Dynamic.html
Best regards,
Eszter
Eszter: Thank you very much. I tried the Auto Local threshold and it might actually be exactly what Im looking for! Thank you for the great advice. I have got a lot to learn but this is extremely helpful.
- Badges
- Science topic
- Similar topics
- Computer Science
- Images