“Graph cut based image segmentation using statistical priors and its application to object detection and thigh CT tissue identification analysis”

Date

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Image segmentation is a field of image analysis that aims to partition an image scene into regions corresponding to objects. It is a popular research topic of image analysis with many applications to the computer vision and medical imaging domains including object recognition and delineation of anatomical structures and tissues. The goal of this thesis is to investigate whether graph cut techniques can be used to delineate the objects in a visual scene for biomedical and computer vision applications. The graph cuts method is one of the leading automated segmentation methods for 2D and 3D images. It delineates the regions by creating graph partitions and finds the optimal graph partition by minimizing an energy function that consists of data and smoothness terms. Graph cuts represent the set of pixels in the image using graph vertices. Relationships between pixels are represented by graph edges and expressed by the smoothness term of the energy function. Source and sink nodes are introduced to the graph to model the region prior information that is used in the data term. An advantage of this method is that it can combine local and global visual information to obtain segmentation of the objects in the visual scene. We perform image segmentation on a database of generic images with reference region masks to illustrate and evaluate the applicability of this method. We applied this technique to generic and medical imaging data for tissue identification. To accomplish that we used two methods one is GC (graph cut) with k means and GC with prior knowledge. We obtained more accurate segmentation results using GC with prior probability (supervised reference masks or polylines) than GC with k-means (automated image segmentation).

Description

Keywords

Citation

DOI

Collections