The ability to automatically extract quantitative data from nonlinear microscopy images is here explored, taking nonlinear and coherent effects into account. Objects of different degrees of complexity were investigated: theoretical images of spherical objects, experimentally collected coherent anti-Stokes Raman scattering images of polystyrene spheres in background-generating agar, well-separated lipid droplets in living yeast cells, and conglomerations of lipid droplets in living C. elegans nematodes. The in linear microscopy useful measure of full width at half-maximum (FWHM) was shown to provide inadequate measures of object size due to the nonlinear density dependence of the signal. Instead, the capability of four state-of-the-art image analysis algorithms was evaluated. Among these, local thresholding was found to be the widest applicable segmentation algorithm.