23-04-2012, 02:29 PM
Wavelet-based Image Compression Using Human Visual System Models
Etdset.pdf (Size: 1.98 MB / Downloads: 107)
Introduction
Motivation
In the emerging digital era, family photo albums are stored not only on the bookshelf, but also
on the personal computer. A medical doctor can make a diagnosis using a full 3-dimensional
image on a computer screen–not long ago surgery would have been necessary to capture the
same critical point of view. Satellite images of earth and places beyond are continually being
transmitted over communication channels. The Internet–still in its childhood–continues
to flourish and impact our personal and professional lives. Common to these and many other
applications is the storage of digital imagery.
Previous Research
Basic wavelet theory and the application of wavelets to image compression has been welldeveloped.
As scalar wavelet filters developed, variations of the scalar wavelet transform were
introduced–among them is the multiwavelet transform [25, 26, 27, 28].
Significance of this Thesis
The application of an HVS model to image compression is a recent research area [17, 18]. The
application of wavelet-based compression schemes to color images has also received limited
attention in the research community. Results have recently been reported for color image
compression using scalar wavelets and an HVS model [17]. Although some of these methods
are similar to those described in this thesis, the results are comparatively minimal and
inconclusive.
Outline of this Thesis
Chapter 2 provides a brief background on transform-based image compression, graycale and
color image compression, and scalar wavelets and multiwavelets. We continue with a short
explanation of the SPIHT quantizer and the modified SPIHT quantization method called
shuffling for multiwavelet decompositions. Chapter 2 concludes with a summary of the
JPEG2000 standard. Chapter 3 provides some background on the human visual system and
the application of this science to digital image compression. Our innovative techniques employed
in the experiments are explained.
Color Image Compression
A digital color image is stored as a three-dimensional array and uses 24 bits to represent each
pixel in its uncompressed form. Each pixel contains a value representing a red ®, green
(G), and blue (B) component scaled between 0 and 255–this format is known as the RGB
format. Image compression schemes first convert the color image from the RGB format to
another color space representation that separates the image information better than RGB.
In this thesis the color images are converted to the luminance (Y), chrominance-blue (Cb),
and chrominance-red (Cr) color space.