158
IRE TRANSACTIONS ON INFORMATION THEORY
December
Another scheme which has been proposed is the run-
length coding, or differential-coordinate encoding system.’
Although this system is applicable to half-tone pictures3
we shall, for simplicity, consider its implementation only
for black and white pictures. The basic idea in this scheme
is to transmit only the lengths of the black and white runs
in a picture as they occur in successive scanning lines.
The first implementation of this system appears to be
that of Treuhaft.4 However, he gives no results con-
cerning the possible reduction in either time or band-
width for a run-length coding type system. Deutsch” has
measured the probability distribution of runs in a section
of typewritten material, and concluded that a saving of
approximately two is possible in a run-length coding
system which uses an optimum code. More extensive
studies of the run-length probability distributions for
typewritten material has been made by Michel’ who
concluded from his findings that theoretically a saving
of approximately ten is possible with a run-length coding
system which employs an optimum code.
These results5e6
are sufficiently encouraging to warrant
further investigation of the run-length coding system. The
present work stems from a desire to predict more generally
the saving that is possible with a run-length coding
system for various types of black and white pictures,
and to gain an insight into this system. The analysis is
carried out only for black and white pictures, and for a
binary digital transmission channel.
TIIE DESCRIPTION OF A PICTURE IN
TERMS
OF A
FIRST-ORDER MARKOF’F PROCIGSS
The
process of scanning reduces a picture from a two-
dimensional array of cells (resolution elements) to a one-
dimensional sequence of cells. In the case of a black and
white picture such a sequence would consist of a succession
of black and white cells. A section of this sequence might
appear as below.
. . .
BBBWWBBBBBBWWWWBWBBWBWW . . .
Thus, in our subsequent discussion, when we use the
word “picture” we shall, in reality, be referring to a one-
dimensional sequence of cells which results from scanning
a picture. (N&z: Since there are many ways to scan a
picture, this representation is not unique. However, for
our purposes, this is unimportant.)
2 A. E. Laemmel, “Coding Processes for Bandwidth Reduction
in Picture Transmission,” Microwave Res. Inst., Polytechnic Inst.
of Brooklyn, Brooklyn, N. Y., Rept. No. R 246-51; August, 1951.
3 W. F. Schreibcr and C. F. Knapp, “TV bandwidth reduction
bv digital coding,”
1958 IRE
NATIONAL CONVENTION RECORD,
pt.
4, pp. 88-99. -’
1 M. A. Treuhaft, “Description of a System for Transmission of
Line Drawings with Bandwidth-Time Compression,” Microwave
Res. Inst., Polytechnic Inst. of Brooklyn, Brooklyn, N. Y., Rept.
No. R 339-53; September 4, 1953.
6 S. Dcutsch,
“Some Statistics Concerning Typewritten or
Printed Material,”
Microwave Res. Inst., Polytechnic Inst. of
Brooklyn, Brooklyn, N. Y., Rept. No. R-526; October RI, 1956.
6 W. S. Michel, “Statistical encoding for text and picture com-
munication,” Conamun. and Electronics, vol. 35, pp. 33-36; March,
1958.
The assumption, on which the analysis presented
herein is based, can now be stated.
The transition in intensity from a given cell to the im-
mediately succeeding cell is determined solely by the
intensity of the given cell.
This dependence is a probabilitistic one which is given
in terms of the transition probabilities pw (w), p,(b),
p,(b), and p6(w), where7
pw(w) is the probability that a cell is white, given that
the immediately preceding cell is white;
p,(b) is the probability that a cell is black, given that
the immediately preceding cell is white
= 1 - pw(w);
pb(D) is the probability that a cell is black, given that
the immediately preceding cell is black;
p*(w) is the probability that a cell is white, given that
the immediately preceding cell is black
= 1 - pa(b).
Two other parameters which enter into the analysis are
are p(w) , and p(b), where’
p(w) is the probability of obtaining a white cell;
p(b) is the probability of obtaining a black cell
= 1 - p(w).
The transition probabilities can be related to each
other, by noting first that
dw; b) = 20; 4
where
p(w; b) = the joint probability of obtaining a white
cell followed by a black cell;
p(b; w) = the joint probability of obtaining a black
cell followed by a white cell.
Hence, since p(w; b) = p(w)p,(b), and p(D; w) = p(b)p,(w),
we obtain
P(Whm = P(bh(W).
It is easily seen that only two of the six probabilities
P(W), p(b), FL(W), FL(~), n(b), pdw) are independent;
for example, an independent set comprises p(w) [or
p(b)], and any one of the four transition probabilities.
At this point we note that the assumption is equivalent
to stating that a picture can be represented by a first-
order Markoff process with stationary transition prob-
abilities.g This model for an information source has been
proposed previously by Oliver,” among others.
f It is well to emphasize that the transition probabilities are not
unique!
since they depend on the particular scanning direction
which 1s used. However! for a given scanning direction the transi-
tion probabilities are unique.
8 p(w) and p(b) are unique for a particular picture, since they
do not depend on the scanning direction.
9 W. Feller, ‘*An Introduction to Probability Theory and Its
Applications,” John Wiley and Sons, Inc., New York, N. Y., vol.
1, 2nd cd.; 1957.
10 B. M. Oliver. “Efficient codina.” Bell Sus. Tech. J.. vol. 31.
pp. 724-750; July,‘1952.
-,