Understanding the content of images is a problem with many applications. Whether it is a robotic system that needs to identify the objects it is instructed to manipulate, a medical system that needs to localize organs or tissues, an image retrieval system designed to return images whose content match a query, or a surveillance system that is expected to identify intruders or watch when objects are moved - all those systems can benefit from a high-quality segmentation algorithm.
Image segmentation methods divide the image into regions of coherent properties in an attempt to identify objects and their parts. Once an image is segmented, the tasks of recognition, compression, and information retrieval can be vastly simplified. In spite of many attempts, existing methods for segmentation fail to achieve satisfactory results when tested on a large variety of natural images. This is due to the vast complexity of images. Regions of interest may differ from surrounding regions by any of a variety of properties, and these differences can be observed in some, but often not in all scales. A further complication is that coarse measurements, for detecting these properties, cannot be obtained by simple geometric averaging, because they would often average over properties of neighbouring segments, making it difficult to identify the segments and to reliably detect their boundaries. Thus this processing, which is easily carried out by a human observer, is extremely sophisticated and non-trivial to achieve with computer algorithms.