estimate for the LPS. Because this subdivision is done approximately, it is possible for the LPS sub-interval to be larger

than the MPS sub-interval. When that happens a "conditional exchange" interchanges the assignment of the sub-intervals

such that the MPS is given the larger sub-interval.

more probable binary decisions can sometimes be coded at a cost of much less than one bit per decision.

be used in the coding of each binary decision. When a given binary decision probability estimate is dependent on a

particular feature or features (the context) already coded, it is "conditioned" on that feature. The conditioning of

probability estimates on previously coded decisions must be identical in encoder and decoder, and therefore can use only

information known to both.

identified by a unique context-index S. The arithmetic coder is adaptive, which means that the probability estimates at

each context-index are developed and maintained by the arithmetic coding system on the basis of prior coding decisions

for that context-index.

X'8000' can be regarded as the decimal value 0.75. The probability interval, A, is kept in the integer

range

to keep C from overflowing a byte of data is removed from the high order bits of the C-register and placed in the

entropy-coded segment.

bytes are stuffed after each X'FF' byte in the entropy-coded segment in order to avoid the accidental generation of

markers in the entropy-coded segment.

the sub-intervals would require:

Qe(S). Whenever the MPS is coded, the code register is left unchanged and the interval is reduced to A Qe(S). The

precision range required for A is then restored, if necessary, by renormalization of both A and C.

make the LPS sub-interval larger than the MPS sub-interval. If, for example, the value of Qe(S) is 0.5 and A is at the

minimum allowed value of 0.75, the approximate scaling gives one-third of the probability interval to the MPS and two-

thirds to the LPS. To avoid this size inversion, conditional exchange is used. The probability interval is subdivided using

the simple approximation, but the MPS and LPS sub-interval assignments are exchanged whenever the LPS sub-interval is

larger than the MPS sub-interval. This MPS/LPS conditional exchange can only occur when a renormalization will be

needed.

identifying the probability estimate used in coding the decision.

estimate for the context currently being coded. No explicit symbol counts are needed for the estimation. The relative

probabilities of renormalization after coding of LPS and MPS provide, by means of a table-based probability estimation

state machine, a direct estimate of the probabilities.