In this talk, I will present a recent work in collaboration with physicists on the analysis of real time Transmission Electron Microscopy (TEM) images to understand molecular transition from crystal solid state to liquid state. Molecules are deposited on graphene with multilayer structures, which are projected and overlaid in noisy 2d TEM images. The problem is to find all the molecular centers in the extremely noisy 2d images where projected molecules are overlaid and to track the centers across the image frames. Before discussing the method that we considered, I will give a brief history in the development of image segmentation techniques with some theoretical and numerical details of old fashioned methods. Then, our method of image segmentation for molecular center identification follows.