Databases Reference
In-Depth Information
2.5
Original
Reconstructed
2.0
1.5
1.0
0.5
0
−0.5
−
1.0
−1.5
−2.0
180
200
220
240
260
280
300
320
F I GU R E 9 . 13
Original 16-bit speech and compressed 3-bit speech sequences.
2.5
Original
Reconstructed
2.0
1.5
1.0
0.5
0
−0.5
−1.0
−1.5
−2.0
−2.5
180
200
220
240
260
280
300
320
F I GU R E 9 . 14
Original 16-bit speech sequence and sequence obtained using an
eight-level forward adaptive quantizer.
distributed over different regions, the range of the input changes. In a forward adaptive
quantization scheme, we would obtain the minimum and maximum values for each block of
data, which would be transmitted as side information. In Figure
9.15
, we see the Sena image
quantized with a block size of 8
8 using 3-bit forward adaptive uniform quantization. The
side information consists of the minimum and maximum values in each block, which require
8 bits each. Therefore, the overhead in this case is
×
16
8
8
or 0.25 bits per pixel, which is quite
small compared to the number of bits per sample used by the quantizer.
The resulting image is hardly distinguishable from the original. Certainly at higher rates,
forward adaptive quantization seems to be a very good alternative.
×
