Active contour models for medical image segmentation

• Main Author: Tiew, Ee Horng.
• Other Authors: Teoh Eam Khwang
• Format: Final Year Project
• Language: English
• Published: 2009
• Subjects: DRNTU::Engineering::Electrical and electronic engineering
• Online Access: http://hdl.handle.net/10356/17895
• Institution: Nanyang Technological University

Image segmentation is a fundamental process in image analysis. It is responsible to partition an image into sub-regions based on a required feature. Active contours model has been a popular methods used as an image segmentation solution because of its ability to produce sub-regions with continuous boundaries. The region-based active contours model is independent of image gradient, thus it produces better segmentation for the image with weak object boundaries. It is also less sensitive to the location of initial contours. The use of level set function has provided the implementation of active contours in a more flexible and convenient way. Specifically, the Chan-Vese active contour model segmentation has been well recognised as an effective approach. It is based on minimizing the energy function that drives the evolution of the contours based on level set. In this project, we investigate and implement the Chan-Vese region-based active contour model. Various synthetic images, real images and MR brain images are being put to experiments and discussions. It is efficient in segmenting objects of disjoint intensities but suffered lost of accuracy in segmenting MR brain images that are more complex in topologies. Also, periodical re-initialized the contours increases the accuracy of the results and reduces the computational time. The implementation was then extended to multiphase segmentation of Vese-Chan active contours model. It is capable to segment more than one objects of different intensity. A 4-phase model was implemented with two level set functions. The multiphase formulation produces no overlap between segmented objects but the computational cost is significantly higher. It provides more dimensions to segment and represent images and inherits the strength and weaknesses of Chan-Vese model. As the challenge for the Chan-Vese active contours and other region-based models is to segment image regions with inhomogeneous intensities, some recommendations for future work will be presented in closing. These include the fusion of local binary fitting (LBF) model and the Chan-Vese model, the parametric shape-based model and the joint curve evolution model.