ADS Technologies USBAV-708 INSTANT VIDEOMPX, User Manual

Page: 27 / 116

The
Guide
27
Fo
rw
a
rd
Qu
a
nt
ize
rs
Quantizers
DivX uses a technique called quantization to control the accuracy of
the image data it stores. Quantizers are similar to the denominator in
an expression such as:
Result =
… and the co-efficient in the inverse quantization:
Data = Quantizer * Result
By maintaining some fixed Quantizer value (greater than one), it is
possible to reduce the value of Data using the first expression, and
then invert the process using the second expression.
However, after quantization the
Result is normally rounded down to the
nearest whole number. This has some interesting consequences for
quantization with regards to image quality.
Examine the following two tables. The first quantizes a series of
number with a fixed quantizer of 3. The second quantizes the same
series of numbers but with a fixed quantizer of 5. Both tables round
down all of the results to the nearest whole number then perform the
inverse quantization and record the error from the original data series.
Data
Quantizer
Data series
Quantization data
Round-down result
Inverse quantization
Error
7
2.33
2
6
14%
14
4.66
4
12
14%
21
7
7
21
0%
28
9.33
9
27
4%
35
11.66
11
33
6%
42
14
14
42
0%
49
16.33
16
48
2%
Quantizer = 3
Data series
Quantization data
Round-down result
Inverse quantization
Error
7
1.4
1
5
29%
14
2.8
2
10
29%
21
4.2
4
20
5%
28
5.6
5
25
11%
35
7
7
35
0%
42
8.4
8
40
5%
49
9.8
9
45
8%
Quantizer = 5