Segmentation of a microscopy image with uneven illumination

Question

I am trying to do a little image segmentation on some cells taken from microscopy images that have uneven illumination. I am having trouble getting the background of the image cleaned up after applying LocalAdaptiveBinarize so I can end up with closed white foreground cells on a plain black background.

Here is the original image:

Due to the uneven illumination (darker in the upper left part of image and brighter in the lower right) I couldn't find a simple threshold that worked well across the whole image so I tried to use LocalAdaptiveBinarize by doing LocalAdaptiveBinarize[image, 10]. Here is the result:

The cells aren't bad but there are a couple annoying defects I would like to improve: One of the two cells in the middle at the top has a chunk of it missing on its lower edge and the cell beneath it has a hole inside it. I can fill in the hole of the bottom cell by using FillingTransform[localBinImage,1] and get this:

I would be interested if anyone has any suggestions on how to improve the cell at the top with the chunk missing from its lower border.

At any rate, with the exception of the top middle cell the rest of the cells now look close to passable but the background is still a mess from using LocalAdaptiveBinarize. I would like to have a way to clean up the background so I just have the cells by themselves on a plain black background that makes it easy to use MorphologicalComponents or something similar to segment the cells.

Any suggestions on how to clean up this background?

Answer 1

A clunky way of doing this, and I can't resolve the smaller cells, but here it goes. Taking from my answer here

img  = Import["https://i.stack.imgur.com/5ni1u.jpg"];
imgd = ImageData@ImageAdjust@ColorConvert[img, "GrayScale"];
imgp = Flatten[
   Table[{x, y, imgd[[x, y]]}, {x, 1, 346}, {y, 1, 462}], {2, 1}];
imgbg = Table[imgd[[x, y]] - nlm[x, y], {x, 346}, {y, 462}]; (* takes a while *)

Here, I am fitting the image data to a plane, assuming that the lighting effect is linear. One may improve the fit with a different function in the NonlinearModelFit if desired. The results can be seen by plotting the ImageData:

Before:

After:

We got rid of the slant, which is what this question is asking for. As for whether or not this improves the cell isolation, I'm not sure:

MorphologicalComponents[ImageConvolve[Image[imgbg], DiskMatrix[1]], 
   0.3] // Colorize // DeleteSmallComponents

It does separate the two cells of interest to the OP, but I couldn't find a decent set of parameters or filters to isolate the two cells to the left or at the bottom of the image. The original microscope image, if it is of a higher resolution, may be needed at this point.

Answer 2

Detect cells:

cellEdges = 
 GradientFilter[ImageAdjust[img], 2] // ImageAdjust // Binarize
cells = Closing[cellEdges, 10]

Separate detected cells:

MorphologicalComponents[cells] // Colorize

Answer 3

I believe that you should use a high-pass filter. The built-in HighpassFilter seems very slow and produces artifacts. Here is an approximate method that is faster and appears to yield a better result:

hp[img_Image, w_Integer] := Image[0.5 + Subtract @@ ImageData /@ {img, Blur[img, w]}]

Application:

img = Import["https://i.stack.imgur.com/5ni1u.jpg"];

img2 = hp[img, 30]

Threshold[img2, 0.52]

p.s. It is shorter to use ImageApply but also slower:

hp2[img_Image, w_Integer] := ImageApply[# - #2 + 0.5 &, {img, Blur[img, w]}]

Answer 4

I guess it depends on exactly what you are looking for, but here is one way to approach this: smooth with a RangeFilter and then Binarize.

img = Import["https://i.stack.imgur.com/5ni1u.jpg"];
Binarize[RangeFilter[img, 1], 0.1]