Sometimes the targets you want to analyze in the images are quite complex in morphology and thus not easily detetcable by their intensity signal. However, Fiji offers the possibility to create a machine learning classifier, which is able to dectet specific objects based on their pixel values. This is a very easy method which does not need any programming skills 
. You just define different object classes (like background, living cells, and dead cells) and mark some of these objects on the images. By the use of the "trainable weka segmentation" the machine learning classifier will learn the differences of these objects and produce a probability map, where the presence of the target objects is encoded as intensity. It is then possible to count the local intensity maxima to count all objects of interest.
More information can be found here: https://imagej.net/plugins/tws/
The original publication can be found here: https://doi.org/10.1093/bioinformatics/btx180
