The determination of the galaxy merger fraction of field galaxies using automatic morphological indices and photometric redshifts is affected by several biases if observational errors are not properly treated. Here, we correct these biases using maximum likelihood techniques. The method takes into account the observational errors to statistically recover the real shape of the bidimensional distribution of galaxies in redshift-asymmetry space, needed to infer the redshift evolution of galaxy merger fraction. We test the method with synthetic catalogs and show its applicability limits. The accuracy of the method depends on catalog characteristics such as the number of sources or the experimental error sizes. We show that the maximum likelihood method recovers the real distribution of galaxies in redshift and asymmetry space even when binning is such that bin sizes approach the size of the observational errors. We provide a step-by-step guide to applying maximum likelihood techniques to recover any one- or two-dimensional distribution subject to observational errors.