First, I can recommend to have a read on the following 2 very good publications to get into co-localization analysis:
Bolte and Cordelières
Dunn et al.
Second, brightness and/or contrast changes should never be done before any measurements related to pixel intensities, since they change the values (as @gabriel stated already). The only manipulation which needs to be done is a background subtraction if there is a considerable amount of background (because this will give you false-postive colocalization indications). Here, Dunn et al. describes an efficient method. Additionally, noise should be as low as possible (e.g. due to line/frame averaging at the microscope, if possible)
… does that mean you take Z-stacks and then do a Z-Projection (combination of Z-slices into a single image)? This could not be used as an input in a co-localization analysis. Rather take the full Z-stack of the 2 channels you want to analyze and start the analysis on them (after background subtraction) by using either JACoP (explanations in the Bolte and Cordelières publication) or Coloc2.
If you have small puncta you are interesteed in the object-based co-localization analysis in JACoP might be a good option and gives you an easy to interpret readout. Also the Manders quotients might be worth looking at. As @ctrueden stated, better use those (statistical) tools because looking for yellow in a combination of a red and a green channel is only in very obvious cases giving a hint on a colocalizing situation.
Once you are familiar with the principles stated in the Bolte and Cordelières paper the Colocalization finder mentioned by @philippe.carl will also be very helpful.