Image Foresting Transform using non-smooth connectivity functions and its applications in the segmentation of medical and natural images with strong heterogeneity

Abstract

Quantitative analysis of brain structures from magnetic resonance images (MRI) has played an important role in research in neurology and can be very useful in diagnosis and treatment of diseases related to changes in the anatomy ofthe human brain. Analysis techniques, however, require a precise definition of the extent of the three-dimensional structures under study. This operation called image segmentation is one of the most fundamental and challenging problems in image processing and computer vision.Methods based on the Image Foresting Transform (IFT) have been successfully used in the segmentation of MRI 1.5 Tesla. However, within the framework of the IFT, there is a class of connectivity functions little explored, which corresponds to the non-smooth connectivity functions (FCNS). These functions are more adaptive and can cope with the problems of field inhomogeneity, which are common in MR images of 3 Tesla. However, further study is needed.The solutions may also be useful in the handling of natural images with strong variation in brightness / shadow. The proposal aims to investigate the FCNS from the standpoint of theoretical and experimental aspects.More specifically, the methods developed will be integrated into a single computational tool that will serve as an important step for several other future research, continuing the work that has been developed under the program-CInAPCe FAPESP, which is held in conjunction with the professors of the Department of Neurology, University of Campinas and other institutions.