Do you have a suggestion to efficiently normalize the area of regions in a binary image created from thresholding?

Dear community,

I have thresholded my way to center-regions of bacterial cells in a bright field image, which I would eventually like to use to measure fluorescence images. My current problem is that my thresholding results in regions of variable sizes, e.g. between 1 and 10 pixels. I would prefer that all regions are 4-6 pixels in size. I would like to add that morphology is a very important factor here, i.e. cells that are elongated have an elongated region, which is something I'd like to retain.

A possible solution i was thinking was, set a desired region size interval, e.g. 4-6, and then in a while loop extract regions below the interval and dilating them, while extracting and eroding regions above the interval until they all fit within the interval. But with my limited coding experience I am guessing there are more robust and less time consuming solutions? Atleast using dilation and erosion with a disk, I would retain morphology overall.

Thank you for your time.

[Guillaume]

I'm curious why you'd want to have a fixed size for your blobs instead of them being proportional to the size of the cells.

[Sebastian Bech-Terkilsen]

Sure. Variable sizes might eventually give bad measurement values if say, a region was slightly too big and covered some non-cell parts or too much of the edges. Therefore, if I was to always have regions of e.g. 4-6 pixels, they are more likely to fit within the centre of the cell and therefore produce higher accuracy. Similarly, a region of 1 pixel might also produce an inaccurate result. Edit: the blobs are not entirely proportional to the size of the cells due to variation in the values that I threshold to.