I followed David's advice to unmix the color and identify cells accordingly. However please note that many cells are overlapped on the top of each other, it's difficult to make precise segmentation. So here in my demo pipeline, it's only a rough estimation.
Regarding cell density, we can try a quick trick that: divide your image 4032 x 3024 into 32x24 grid, each tile of the grid sized 126 x 126 pixel.
Then you do relate object between the cells and the tile (tile is parent, cells are children), then count how many cells per (126 x 126) square.
As Beth already pointed out, if you want precise cell/mm2, you need to know how big is 1 pixel in mm, then simply convert 126 x 126 pixels into mm.
pipeline_2.cpproj (641.8 KB)
Hope that helps