
i
i
i
i
i
i
i
i
168 8 Directional DCT Transform
(a) (b)
Figure 8.8 Direction modes selected in Barbara with 1.07 bpp and in Lena with 0.69 bpp.
from JPEG, but the first 1D transform is replaced by the proposed directional DCT-
like transform. The values of fractional pixels needed are calculated by the 8-tap Sinc
filter. The direction of a block is first predicted from the coded directions of neigh-
boring blocks and then is coded by the Huffman coding losslessly. Therefore, JPEG
is selected as an anchor for comparisons.
We report the experimental results of eight common images: Barbara (512×512),
Lena (512 ×512), Peppers (512 ×512), Photographer (256 ×256),