Multimedia & Applications

Biography

Biography: Abdellatief Hussein A Ali

Abstract

Vector quantization, VQ, is the process of picking suitable set of representative vectors, code book, to an extremely larger set. The use of representatives or their indices was widely applied in communication and multimedia compression. VQ techniques, for example k-means, adapt the code book vectors location in space to minimize representative’s overall bin distortion. The VQ techniques treat boundary vectors contribution to distortion as any other point that leads to the possibility of forming clusters across the boundary of the classes or objects. The study highlights the possibility of having representatives that observe classes or objects boundaries. The code book generated is out of learning process to the classes or the connected regions. Consequently, the code book is not only effective in compression but also suitable for applications such as classifications and recognition. The proposed code book generation process is composed of three phases: initialization, iterative, and finalization. In the initialization, the min-max algorithm is used to pick the initial code book. The min-max algorithm enables the choice of the code book size based on ceiling to the expected distortion. In the second phase, adapted LBG iterates on the code book, starting with the initial, to learn classes or objects boundaries. The iterative process focuses on using the miss-represented points to attract the right representative and repels the current. In the finalization phase, code book point that does not contribute to correct decoding is dropped from consideration in the final code book.