Date of Award

Spring 2005

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Electrical & Computer Engineering

Program/Concentration

Electrical Engineering

Committee Director

Vijayan K. Asari

Committee Member

Stephen A. Zahorian

Committee Member

Min Song

Call Number for Print

Special Collections LD4331.E55 B69 2005

Abstract

Image compression is used to remove different types of redundancies and irrelevant information in an image so that the space required for storing the image is minimized. Conventional transform based techniques remove the redundancies between adjacent pixels and irrelevant information in the image but fail to remove the redundancies between different parts of the image. There are techniques like vector quantization, which take care of the redundant information in different parts of the image, but fail to remove the irrelevant information in the image. In this thesis a new technique for lossy digital image compression is proposed which is a combination of both transform based compression and vector quantization. The algorithm presented in this thesis not only removes the redundancy between neighboring pixels and the irrelevant information present in the image but also the redundancy between different parts of the image. The compression ratio attained by this algorithm is better compared to both transform based and vector quantization based compression techniques.

The new image compression technique first uses Discrete Cosine Transform (DCT) to separate the high frequency information. This high frequency information is removed completely or quantized heavily depending on the compression required. A self-organizing neural network based technique in combination with vector quantization is used to detect different parts of the image that are similar. These blocks of the image, which are similar, are classified and represented using a single entry in the codebook. A first order predictor method instead of a conventional zero order predictor has been used to encode the index. The first order predictor takes advantage of the fact that the change of gradient in an image is smooth. The change of gradient is assumed to follow the direction with minimum change in gradient and any change in the value of the current block from the predicted value is stored in the index. The results of the experiments conducted using this algorithm show an increase in performance by approximately 20% to 80% (in terms of compression) when compared to the JPEG standard. When compared to JPEG 2000, the proposed algorithm yields better compression at higher rates. Research work is in progress to improve the performance of the algorithm by incorporating an optimum quantization matrix and an adaptively computed vigilance value for the self-organizing neural network.

Rights

In Copyright. URI: http://rightsstatements.org/vocab/InC/1.0/ This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).

DOI

10.25777/t2hd-fv31

Download

Share

COinS