Department of Pattern Recognition

Abstract: -

Image segmentation is fundamental to many medical image analysis tasks. Some examples are tumour lesion delineation for radiotherapy planning, liver segmentation on CT for liver transplant planning and cancer detection in breast screen, lung screen and virtual colonoscopy. There are many different approaches in segmenting an image. For example, probabilistic atlas, level-sets, graph-cuts and rule-based systems have been explored for liver segmentation. The challenge is in multi-class segmentation including not only solid organs such as heart, liver and kidneys but also tissues like muscle and fat. In this presentation, we introduce the superpixel approach which reduces the image to a collection of superpixels. These superpixels are regions ‘homogeneous’ with respect to certain property of the image. Criteria for the superpixels and techniques in attaining the superpixles will be discussed.