Segmentation of multiple objects evolving conditional random field based topology adaptive active membrane

In this paper we have used conditional random field based learning scheme to differentiate the spectral signature of the objects and background in a scene. The overall objective is to segment multiple objects in a poorly contrasted scene. The primary tool for segmentation is a region based active membrane which evolves under image based external energy. The learning scheme helps in splitting the active membrane for segmenting multiple objects and integrates the topology adaptive property of the active membrane with the architecture and evolution of the membrane. The proposed approach is tested in a challenging application domain of estimation of sizes of oil sand rocks.

► Segmentation is achieved for low contrast touching objects. ► CRF based learning scheme is unified with active membrane evolution. ► Learning helps in preserving topology adaptive property of membrane. ► Images of oil-sand rocks and blood cells are successfully segmented.