Canon FACSIMILE BASIC 2000 Service Manual page 180

G3 FACSIMILE COMMUNICATIONS
The concept of alithmetic encoding is simply explained below.
The following assumptions are made in order to make the explanation easy
to understand.
The special characteristic of this arithmetic encoding is that an additional
encoding bit is not needed because the integer line position is the same as
the integer line position of the previous encoding data in the case of an
accurate prediction. It follows that the amount of encoding will not
increase if accurate predictions continue, and the rate of compression will
increase. Conversely, with inaccurate predictions, an additional encoding
bit will be necessary to show the position of the inaccuracy in detail, and
thus the amount of encoding will increase and the rate of compression
decrease. In this way, the study table learns in order to increase the rate of
accurate predictions and to reduce the amount of encoding and raise the
compression rate during the encoding process, and then corrects the table
parameters.
*1 The actual probability varies with the status because of the extent to
*2 The actual range is hexadecimal 8000 to 10000.
5–38
more detail. This operation is called "Renormalization", and this range
(A) is reset to a value above a certain range
The probability of accuracy will be 50%, and the probability of inac-
curacy will be 50%.
The area of accuracy will be MPS, and the area of inaccuracy will be
LPS.
Position (Binary)
1.000
0.111
0.110
LPS
0.101 Range
A1
0.100
0.011
LPS
0.010
MPS
0.001
MPS
0
Pixel encoded
1st 2nd
Accuracy
Accuracy
Each pixel's code
0 0
position according to
arithmetic encoding
Fig. 5-31 Arithmetic Encoding Conceptual Diagram
which LSZ occupies in the range (A).
In the case of an accurate prediction, range A1= hexadecimal 10000-
*1
0.0011
Range
A2 Range
A3
LPS
LPS
MPS
MPS
0.0010
3rd 4th
Inaccuracy
Accuracy
0.001
*3
.
LPS
0.001010
0.001001
LPS
0.00100010
MPS
LPS
LPS
MPS
MPS
MPS
5th 6th 7th 8th
Inaccuracy
Accuracy
Accuracy
Accuracy
0.001 0.00100010
0.001
0.00100010