Color image segmentation using Multi-Scale Clustering

Due to the character of the original source materials and the nature of batch digitization, quality control issues may be present in this document. Please report any quality issues you encounter to digital@library.tamu.edu, referencing the URI of the item.Includes bibliographical references: p. 72-74.Image segmentation is the partitioning of an image raphics. into homogeneous regions. Colors, intensities, textures, edges, or other relevant features may dictate the nature of the homogeneities. Usually the homogeneous regions represent objects of interest. Segmentation is the first step in applications such as object recognition and tracking. This research concerns color image segmentation. Color image segmentation is the partitioning of an image into homogeneous color regions. Often color provides a distinguishing feature for objects of interest. Most current, color image segmentation methods suffer from two major deficiencies. First, the majority of segmentation procedures are not objective. That is, they require subjective, user-defined criteria to determine the segmentation results. Secondly, the methods are dependent upon initial conditions. Essentially, their solutions are local, not global, optimizations. This thesis proposes the use of an algorithm called Multi-scale Clustering (MSC) for color image segmentation. MSC is able to objectively determine the prominent colors in an image independent of initial conditions. Additionally, the MSC segmentation process can be applied to both edge detection and compression. MSC, like all clustering algorithms assumes the clustering space is uniform. Unfortunately, the majority of popular color spaces are non-uniform. This thesis explores the effects of clustering in both uniform and non-uniform color spaces