Improving 2D mesh image segmentation with Markovian Random Fields
Alex J. Cuadros-Vargas, Leandro Cavaleri Gerhardinger, Mário de Castro, João Batista Neto, Luís Gustavo Nonato · 2006
Traditional mesh segmentation methods normally operate on geometrical models with no image information. On the other hand, 2D image-based mesh generation and segmentation counterparts, such as Imesh (A. Cuadros-Vargas et. al, 2005) perform the task by following a set of well defined rules derived from the geometry of the triangles, but with no statistical information of the mesh elements. This paper presents a novel segmentation method that combines the original Imesh image-based segmentation approach with Markovian random field (MRF) models. It takes an image as input, generate a mesh of triangles and, by treating the mesh as a Markovian field, produces quality unsupervised segmentation. The results have demonstrated that the method not only provides better segmentation than that of original Imesh, but is also capable of producing MRF-like segmentation output for certain types of images, with considerable cut in processing times