How to seperate a cluster and get count result

Hello everyone!! I have been stuck in cell-counting for a long time.I try to use imagej to count signal numbers based on pixels.There is the procedure:
1.colour deconvolution
2.gaussian blur
3.set threshold
4.analyse particles
5.count roi(=signal numbers)

But some signals will be recognized as one signal because of their close distance.
original photo:
So the result will be less than the actual numbers. I really can’t figure out this.Could anyone help me?Thank you so much!!

I do not think this can be resolved in a robust way from that image alone.
If you could get a clean image of the nuclei only, then you could try defining the cell population, then expand the nuclei mask (without merging) and then detect which are the cells which are “positive” by looking at the distribution of the intensity of the pixel inside the expanded mask. That is, you need the nuclei to have an idea of which cytoplasm belongs to which cell.
Hope this gives you some ideas.