Automated cell counting on fluorescent stained bacteria images is a rapid and accurate method for microbial quantification, that identifies and counts individual cells, also sorting them within different subpopulations. The key process is the image binarization, where cells (white pixels) are identified within a homogenous background (black pixels). We make an extensive analysis to evaluate the performance of 25 binarization approaches over 16 synthetic images of bacteria with two different morphologies (cocci and bacilli), and we find that the global thresholding methods Intermode, IsoData, Moments and Otsu slightly outperforms the others. Since confocal images are often non uniformly illuminated, we show that some local thresholding algorithms are more efficient in cell counting. Local Bernsen, Otsu and Phansalkar methods achieve the best performances, if the region of interest (ROI) is suitably selected. We perform a detailed analysis of Bernsen algorithm and theoretically demonstrate that its performances can be enhanced by a suitable selection of the ROI dimension and the user-defined threshold parameter.