Genetic based clustering for image segmentation and vector quantization.

number: 
489
English
department: 
Degree: 
Imprint: 
Computer Science
Author: 
Basher Jassim Hamad Al-Ani
Supervisor: 
Dr.Layla S. Al-Ali
year: 
2000
Abstract:

The purpose of this research is to see how useful it to use genetic programming in performing image clustering as well as in performing vector quantization for image compression applications. Two genetic based clustering algorithms (Centroid Based Genetic Clustering Algorithm CBGCA and Partition Based Genetic Clustering Algorithm CBGCA) implemented for clustering a two-dimensional gray level images. Also, the traditional K-means clustering algorithm implemented, which used for comparison purpose. Results from the use of the suggested genetic based clustering methods (CBGCA, PBGCA) and results from the classical clustering method (k-means) were compared, and acceptable results are obtained. The hybridization between the genetic based clustering methods (CBGCA, PBGCA) and the traditional k-means suggested, and the experiments have showed that the suggested modification (hybridization) have improved the performance of the genetic based clustering methods and the classical k-means and gave encouraging results. Another two genetic based clustering algorithms (Codebook Based for Vector Quantization using Genetic Aligorthm CBVQGA and partition Based for Vector Quantization using Genetic Algorithm PBVQGA) implemented to compress a two-dimensional gray level images. Also, simple vector quantization system with LBG algorithm for codebook generation implemented for comparison purpose. A comparison between the results obtained from the suggested methods and the classical methods were performed. CBVQGA provides good results and little better then that of the simple vector quantization system with LBG algorithm, but the other suggested method PBVQGA was produce unacceptable results.