The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
247
2. Design the directional filters, for this paper we select the
directional db5/3 filters.
3. Performs the directional decomposition using the images: LH,
HL and HH. The process is made using the usual 5/3 tap filter
that decomposes an image with a maximum of 3 directions or 8
subbands at the finer wavelet subbands.
4. Repeat the step three with the next wavelet level of LH, HL
and HH images, in this paper, we only do the directional
decomposition to the two high-frequency levels.
In Figure 4. We show the process to compute the WBCT (LiYu,
Lin, 2007)..
Figure 4. The process of WBCT
2.4 The Bitplane Coding Based On Significant Bits
Choose an adaptive encode method is very important for image
compression, in this paper, we use the bitplane coding based on
significant bits. For this method is very simple compared with
SPIHT or EZW and time-saving. The matter is how to encode
the bits so we can get a higher compression rate.
Before coding, two conceptions should be referred to (FeiPeng
Li, 2003): the significant bit and the refinement bit. The
significant bit is the first non-zero bit of a coefficient; the bits
after the significant bit are called refinement bits. And the bits
before the significant bit are called zero bits. For zero bits and
significant bits, we encode them with a Multiple Quantization
(MQ) coder, the sign bits are encoded by another MQ coder,
but for refinement bits, we don’t encode them but preserve
directly for each one’s probability of 0 or 1 is equal. The
definition of the three kinds of bits and their encode method are
shown in Figure 5.
Figure 5. Coefficient bits encoding method (shadows are the
significant bits)
2.5 Structure Similarity
As we all know, the classical objective measure PSNR is not
the only evaluation method of image quality, and it is not
appropriated to measure the goodness of our coder. In this paper,
we try to use a structure concerned method to evaluate the
reconstructed images: the Structure SIMilarity (SSIM),
represented by Zhou Wang (Zhou Wang et al. 2005). This
method considered not just the error of each pixel between two
images, but three different factors that decide the image quality:
luminance, contrast and structure. So we choose SSIM as the
image quality assessment method in this paper. The SSIM value
has a constraint from 0 to 1, where high quality image has a
larger SSIM value than a lower quality image.
3. TESTS, RESULTS AND DISCUSSION
In this paper, the experiments are based on the ImagZip2.0
system constructed by doctor Li FeiPeng et al. 2001. In order to
best illustrate the detail preserving effect of our method, we
choose the 512*512 size image”Barbara” which is abundant in
directional details. We use this image to perform two different
tests, the first one with a bit rate of 0.5 in order to obtain a
compression factor of 16, and the second one with a bit rate of
0.2 in order to obtain a compression factor of 40.
3.1 The comparing of the reconstructed images
The decompressed images obtained for test (bit rate of 0.5 and
0.2) using the DWT and the WBCT are shown in Figure 6.
(e) (f)