Calcium imaging of neurons is a powerful tool to measure in vivo neuronal activity where one of the key challenges for calcium spike train characterization is its requirement for rapid temporal sampling to resolve fluorescent transients. However, to distinguish nearby cells that have only small differences in their firing patterns remains a challenge. Current automated image segmentation techniques which rely on the spatial analysis of image intensity, fail to accurately identify single-cell regions of interest (ROI). We present a cell parsing technique that combines a region-growing method and correlation analysis of pixel stacks to find and optimize the ROI boundaries of single cells in calcium imaging times series. While a spatio-temporal correlation function gives a signature of the calcium activity, a region-growing method gathers correlated activity into ROIs. In computer simulations of calcium imaging data sets, the algorithm was able to generate appropriate ROIs including those with irregular morphologies. The algorithm was also applied to in vivo 2-photon calcium imaging of Oregon Green BAPTA-1 in the Xenopus laevis optic tectum and retina, in both cases effectively identifying active single cell ROIs corresponding to confirmed cell boundaries. These results demonstrate that the combination of a region-growing method with correlation analysis for segmentation permits robust automated cell segmentation which has the potential to identify and delineate active neurons permitting automatic analysis of activity in neurons via in vivo calcium imaging.