X-ray Image Segmentation using Active Shape Models

Image segmentation is an important stage in any image processing process. Regions of interest in the image are extracted from the image and are used to interpret the information in the image. This study investigates the image segmentation of X-ray images and aims at separating the bone from the rest of the X-ray. Basic edge detection techniques and Active Shape Models are the image segmentation techniques analyzed. The performance of these methods is tested using error functions and bestt curves. The process is modi ed to automate it to decrease human involvement. These are the initial steps of an automatic bone fracture detection algorithm. Basic edge detection techniques consist of texture analysis and morphological operations on the image to nd edges. In this thesis, the bone can be separated from the X-ray image using the bone boundary. The bone boundary can be found using edge detection techniques. The advantages of these techniques are examined and their drawbacks are explained. A study of these techniques shows their inadequacies in detecting the edges in certain bones, creating the need to nd a robust and e cient way of nding bone boundaries. Active Shape Models, presented in Cootes and Taylor [9, 10], is a method of nding a shape in an image. Active shape models are used to t a shape, learnt from training images, to a test image. The algorithm is trained using X-ray images by manually selecting landmark points on the images. The shape of the bone is learnt using these images and then the model tried to t the shape to a test image. Performance of these models is tested and variations on the model are studied. Interpretation of the results of experiments shows the best way to use Active Shape Models to segment X-ray images. It is proved that this technique extracts the bone in the X-ray e ectively.