The Fusion of Supervised and Unsupervised Techniques for Segmentation of Abnormal Regions

Abstract

Image segmentation is an essential component of any algorithm used for the automatic detection of abnormalities in digital mammograms. Most of the segmentation techniques used by the researchers in the field however use very simple measures such as grey-level values to group similar pixels on the image together. Some researchers have argued that using such simple measures will result in the segmentation of many false positive regions [1]. They have suggested that the use of more sophisticated texture properties will result in the elimination of many of these false positive regions. In addition, these texture properties can also be used to pre-classify some of the abnormal regions.