Original
This image was constructed in an attempt to find the best and worst
of the various compression methods. For example, DXTC does not like images
where localised 4x4 blocks contain diverse colours (as in the noise and stripes),
but handles simple colour changes (eg the edges of the text) and gradual blends
extremely well.
VQ2x2, on the other hand, is is usually quite good at handling high frequency data (provided there is some self-similarity in the image), but often has difficulty with subtle blends or where the image contains a lot of different codes (as is the case with the noise).
The uncompressed image would require 192k at 24bpp, and 128k at 16bpp
Compression Results:
Equal Resolution (256x256)
8 Bit palettised Size: 64k (+ 0.75k table) RMS error: 12.0 |
4 Bit palettised Size: 32k (+ 0.05k table) RMS error: 64.1 |
DXTC Size: 32k RMS Error: 83.4 |
2x2 VQ Size: 16k (+ 2k table) RMS Error: 67.8 |
Equal Storage Space (approx 16k)
(24bit, 8bit, DXTC, and VQ tested)
Basic 24 bpp texture Initial Resolution: 73x73 RMS Error: 124.1 |
8 Bit palettised Initial Resolution: 128x128 RMS Error: 126.1 |
DXTC Initial Resolution: 181x181 RMS Error: 125.6 |
2x2 VQ Resolution: 256x256 RMS Error: 67.8 |