What if I crop the picture after calibration and then measure the particles?

Hi there! I have a question about measuring the size of segmented particles after cropping the calibrated image.

What I have now are several images that are calibrated with the cameraParameters I generated, which contains the calibration coordinates like image 1 below (the right yellowish squares). What I do after the calibration is to crop the image (use imcrop), generate a mask for the particles and measure the sizes of them (the colonies), like image 2 below. So my question is: after I crop the calibrated image into another, are the sizes of the particles I measured also calibrated? If not, how can I maintain the calibrated size after cropping? Thank you!

Image 1

Image 2

Answer 1

[Image Analyst]

You could make your life, and your program easier, by simply fixing your bad image capture environment. Simply make a jig. You should have a light booth with controlled lighting, but at the very least you could put the camera on a copy stand with an L bracket on the floor. Simply shove your dish into the corner of the L and your dish will be perfectly aligned every time. Then you can use fixed rows and columns to crop your image or define a circular ROI. No need to automatically find the dish (with something like stdfilt) when this simple solution will eliminate the need to do that.

[Ziming Zhong]

Thank Image Analyst! I already have a copy stand for the camera and the light, and also I draw a circle on the background to align the dish. I use the light, camera, the background and the cropping coordinators every time the same, in order to make the photographing condition as similar as I can every time. But one thing I'm still concerned is that it's difficult to align the camera body perfectly horizontal every time, which will result in a slightly different angle when photographing. I think some of my colonies are very small, and small difference in photographing angle may make difference. That's why I am thinking about to calibrate the image.

But I'm quite new in image analysis and I'm not really sure about this. What do you think? I already finished all the codes for image analysis, and till yesterday my colleague said I should calibrate the image to make sure every time the measurement is the same. If it is not so difficult, I would like to try.

[Image Analyst]

Absolutely you should do both a spatial and color calibration. Not only that but you should do a background correction to compensate for lens shading and illumination spatial variation. You should also have an enclosed light booth that is not subject to ambient lighting and colors that your clothing may cast onto the sample.

Conceptually it's not difficult but I think you would think all that is difficult to program, so I guess you aren't going to try. Nonetheless, see my attached presentation.

[Ziming Zhong]

Thanks, Image Analyst! I think I need to try this color calibration, after I read through your presentation. You scared me now, as I'm going to do the quantitative analysis for my colony color. But before that, the camera calibration and spatial calibration is more important to me, as it affects my measurements of colony size.

(1) I have read through your demo for the spatial calibration, and I want to calculate the calibration factor, by defining the distance between each cross (right side of the image with little squares) as 1 mm and then dividing 1mm with the pixel distance between the crosses. To do so, I tried detectHarrisFeatures() and detectMinEigenFeatures() to detect the cross, but the function can't detect any cross in my image. Do you know how to do that?

(2) Also, I tried to retake my images with checkerboard, in order to use the camera calibration app to calibrate my image. But I found it hard for me to think out how to combine the cameraParameters and extrinsic parameters (R and t) with `bwlabel` and `regionprop` functions. Can I write something like below?

KeeperComponentsMeasurement=regionprops(keeperComponents,'Centroid','PixelIdxList','Area');
worldPoints1 = pointsToWorld(cameraParams, R, t, KeeperComponentsMeasurement);

Plus, when I run

[R, t] = extrinsics(imagePoints, worldPoints, cameraParams);

I always got an error: Undefined function or variable 'worldPoints'. I don't understand why. Do you have any idea about it?

Thanks!!!!!

[Image Analyst]

I haven't used the camera calibration app so I can't help you there. The regionprops() call looks correct.

I recommend you just set up a rigid jig so that the calibration can be measured just once, like from a black and white ruler or something so you don't have to find it from every image from some faint gridlines.

Or else you could build my spatial calibration stuff (drawing the line and asking the distance, etc.) into a pulldown menu or push button on your GUI.

[Ziming Zhong]

Oh, the ruler idea is great! But I'm thinking even I use a ruler in the image, I still need to manually measure the pixel distance between two grids on the ruler, which create errors every time (because pixels are so tiny and I can't even get the same pixel coordinates if I manually crop the image).

I already set up a rigid jig and marked the place for plates. But I need many days to take pictures and the camera was removed and reset on the jig in different photographing days, which leads to the major error in manually measure the pixel distance. How can I solve this problem? Or the error in pixel distance is too small to take into account?

[Image Analyst]

You might need better mounting brackets. Try to get some from Manfrotto or Bogen or RAMMounts. If you can get the camera and lens back into the same position (working distance = lens to sample distance, and same focal length), then the spatial calibration should be the same. So if it's 20 cm, then if you remove and replace the camera it should still be 20 cm field of view, even if it's rotated slightly. If you do want to calibrate off each image then don't use graph paper. Take an object of known dimensions, like a paint chip sample from the hardware store or a metal scale/ruler of known width and measure that in the image.