
5.4 Bit Allocation in Transform Compression
161
5.4
BIT ALLOCATION IN TRANSFORM COMPRESSION
In our discussion so far, each transform coefficient was normalized by
its standard deviation, then the same quantizer was used for all the normal-
ized coefficients, and finally equal length binary code words were used for
representing the quantizer output. With this approach to bit allocation, 8-
bit/pixel image data can usually be compressed to 2 bits/pixel without
excessive visual degradation of the image.
The above approach to bit assignment, although easy to implement, is
not optimal for two reasons, and higher compression ratios can be obtained ...