Wavelet Transform Compression
Our group also implemented a coder based on a wavelet transform. The coder
compresses 256*256 images. The Daubechies wavelet family was used. Both the
Haar and the Shannon scaling functions, are rather impractical for
multiresoloution analyses. The Daubechies family reaches a compromise
between compact support and smoothness. We used the Rice wavelet toolbox to
implement the transform.
Scaling function for a Daubechies wavelet
The coder applies the wavelet transform to the entire image. The transform
contains a collection of sub-images that are scaled versions of the
original image. The values obtained by the transform are scaled so as to be
represented by 8 bits. This includes 1 bit for the sign. The values range
from -127 to 127. The number of bits for each value is determined according
to a specified bit mask. These bits are then written to a file along with mask information.
The inverse coder takes the quantized transform values from the file and
applies the inverse wavelet transform. The image obtained has a quality
proportional to the number of bits assigned per pixel in the bit mask.
Back..