# Optimal Binarize value for counting experiment

I am currently working on a physics experiment where I count particles. I have looked at and used a bit of code from the following sources: Cell counting from an image file and Count Elements in Image

However,this is my first attempt at the code.

(Note: this code wasn't giving good counts for the particles at particular slide depths. I attached my final solution to the end of this post):

locate = SystemDialogInput["FileOpen"];
img = Import[locate]
b = FillingTransform[ColorNegate[Binarize[img]]]
disT = DistanceTransform[b, Padding -> 0];
Method -> "Rainfall"];
cells = SelectComponents[w, "Count",
0 < # < 500 &];(*Pixel interval*)measures =
ComponentMeasurements[
Grid[{{Style[
"Number of Particles is " <> ToString[Dimensions[measures][[1]]] <>
".", "Title"],
Button["Save it?",
Export[DirectoryName[locate] <> FileBaseName[locate] <> ".txt",
"Number of Cell is " <> ToString[Dimensions[measures][[1]]] <>
".     " <> DateString[]]]}, {Show[img,
Graphics[{Red, Circle @@ # & /@ (measures[[All, 2, 1 ;; 2]]),
Text, {measures[[All, 2, 3]], measures[[All, 2, 1]]}]}]]}}]


isn't working very well with my images.

Furthermore, I am having to adjust the value of x in Binarize[img, x] for each image in order to get accurate counts. I don't exactly know how to adjust Binarize to get consistent counts for my data

The final picture in my data set is also proving to be a challenge to process.

I solved the ImageCrop[] problem of this question by pre-processing all of my images, now I just need to find a way to make Binarize[] give good counts of the particles in my image(s).

Final Code

For those who need it.

(The above method gave somewhat inaccurate results so I settled on this one, it basically searches for the images called ker, ker2, and ker3 and finds the points associated with these images in the file that you input to the function. It produced some nice counts)

(ker) = Kernal 1 = (ker2) = Kernal 2 = (ker3) = Kernal 3 =

ker = (*Image of Kernal 1*)
img = Import[locate];
i = ImageCorrelate[img, ker, NormalizedSquaredEuclideanDistance];
dots = Point[#[[2]]] & /@
ComponentMeasurements[
MorphologicalComponents[ColorNegate[Binarize[i, 0.22]]],
"Centroid"];
Length[dots]
Show[img, Graphics[{Red, dots}]]

ker2 = (*Image of Kernal 2*)
i = ImageCorrelate[img, ker2, NormalizedSquaredEuclideanDistance];
dots2 = Point[#[[2]]] & /@
ComponentMeasurements[
MorphologicalComponents[ColorNegate[Binarize[i, 0.22]]],
"Centroid"];
Length[dots2]
Show[img, Graphics[{Red, dots2}]]

ker3 = (*Image of Kernal 3*)
i = ImageCorrelate[img, ker3, NormalizedSquaredEuclideanDistance];
dots3 = Point[#[[2]]] & /@
ComponentMeasurements[
MorphologicalComponents[ColorNegate[Binarize[i, 0.22]]],
"Centroid"];
Length[dots3]
Show[img, Graphics[{Red, dots2}]]
Length[dots] + Length[dots2] + Length[dots3]


You may want to do some processing on the image before your Binarize it.

Working with the Blue channel of your jpg:

img = ColorSeparate[Import["tRGTt.jpg", "JPG"]][[2]];


Use a BottomHatTransform to correct the background

img2 = Binarize[BottomHatTransform[img, DiskMatrix[15]]]


There are many ComponentMeasurements options you can play with to be selective about the components it picks.

morph = MorphologicalComponents[img2, CornerNeighbors -> True];
comps = ComponentMeasurements[morph, {"Centroid", "EquivalentDiskRadius", "Circularity"}, ((5 < #2 < 30) && (#3 > 0.2)) &];

Show[Image[img, "Byte", ImageSize -> Large], Graphics[{Red, Map[Circle[#, 15] &, comps[[1 ;;, 2, 1]]]}]]


It's definitely not perfect, there are some false positives and it misses a lot of your faint, out of focus particles, but hopefully its enough to give you a start! In your measurement, is it possible to keep your particles in the focal plane of your imaging system to make your processing easier? For example by making a thinner flow cell for your particles?

• That looks like it works great for the faint particles. But I realized that I was missing the point of this counting experiment. I was trying to count particles that were out of focus - aka not on the depth of the slide that i was concerned with. – Daniel Schulze Sep 27 '15 at 19:25
• So what I did was use images which related to "In-focus" particles and counted those using the method that matched the pictures of in-focus particles to the picture of the input slide – Daniel Schulze Sep 27 '15 at 19:27