Image segmentation, defined as the separation of the image into regions, is one of the first steps leading to image analysis and interpretation. The goal is to separate the image into regions that are meaningful for a specific task. This may, for instance, involve the detection of organs such as the heart, the liver, or the lungs from MR or CT images. Other applications may require the calculation of white and gray matter volumes in MR brain images, the labeling of deep brain structures such as the thalamus or the hippocampus, or quantitative measurements made from ultrasound images. Image segmentation approaches can be classified according to both the features and the type of technique used. Features include pixel intensities, gradient magnitudes, or measures of texture. Segmentation techniques applied to these features can be broadly classified into one of three groups : region-based, edge-based, or classification. Typically, region-based and edge-based segmentation techniques exploit, respectively, within-region similarities and between-region differences between features, whereas a classification technique assigns class labels to individual pixels or voxels based on feature values.
Because of issues such as spatial resolution, poor contrast, ill-defined boundaries, noise, or acquisition artifacts, segmentation is a difficult task and it is illusory to believe that it can be achieved by using gray-level information alone. A priori knowledge has to be used in the process, and so-called low-level processing algorithms have to cooperate with higher level techniques such as deformable and active models or atlas-based methods. These high-level techniques are described in great detail in Chapters 3 and 17 of this handbook and they will only be mentioned briefly in this chapter to provide the appropriate links. This chapter focuses on the segmentation of the image into regions based only on gray-level information. Methods relying on a single image, also called mono-modality methods, will be presented as well as multi-modal methods that take advantage of several co-registered images (see Chapter 8). Because image segmentation is a broad field that can only be touched upon in a single chapter, the reader will be introduced to basic segmentation methods. A number of pointers to the pertinent literature will be provided for more detailed descriptions of these methods or for more advanced techniques that build upon the concepts being introduced.
Online access to SPIE eBooks is limited to subscribing institutions.