Abstract:
In the wake of increased use of multimedia applications for day-to-day normal activities, including
business and entertainment, the pressure of introducing a generalized technique for segmenting an
object in effective and efficient way becomes crucial for creative animations and providing low bit
rate communications. All the existing techniques of image segmentation either require prior
knowledge about the number of objects in the image or misclassify pixels to be included within the
object or the both. The above limitations, along with the complexities of the existing segmentation
techniques, make the process ineffective as the requirements of segmentation cannot be met in
many cases. This thesis proposes a novel image segmentation algorithm called the robust object
segmentation based on pattern matching, which introduces the region stability, connectivity and
pattern matching techniques for splitting and merging an image to identify objects within it. The
stability test classified the regions as the background, object and mixed regions, of which the most
interesting mixed regions will be splitted recursively until a small block containing 16xl6 pixels is
obtained, otherwise being classified to be either the background or the object region. Each of the
smallest mixed regions is then matched with one of the suitable patterns for replacing the contour of
an object. The algorithm for the first time solves key problems of segmentation, is able to segment
all types of known and unknown images and provides the superior performances compared with
recently developed suppressed fuzzy c-means and object based segmentation using fuzzy clustering
schemes in terms of its complexity and ability to segment objects, and therefore, hastening the
arrival of new era in the world of multimedia applications.
