X-ray transmission videography is used to observe the thermodynamic phase behavior of heavy oil mixtures of industrial interest that are opaque to visible light. The knowledge of the volumes of various phases provide important information for enhancing oil refining technology. Phase boundaries are usually marked by visual examination based upon which the volumes and densities of the phases are computed. Typically, phase behavior is computed at various temperatures and pressures to generate the phase diagram. This paper presents an image processing algorithm, which automates the task of detecting phase boundaries in the generated images. The interface and region models are first obtained from the image generation process. Using these models, a new method for interface detection based on the Mumford-Shah model is presented. Results are presented to show that the algorithm detects the phase boundaries to subpixel accuracy. © 2001-2012 IEEE.