 |
Description  |
|
|
BACKGROUND OF THE INVENTION
The present invention relates to a method for transmitting video signals
with the aid of differential pulse code modulation (DPCM) and a controlled
quantizer, in which the system is switched between different quantizing
characteristics.
For the economical transmission of video signals, such as, for example,
television signals, various types of DPCM method are already in use. The
basic block circuit diagram for such a DPCM system is shown in FIG. 1. An
analog video signal of a limited bandwidth from a picture source, which
may be a camera, transparency scanner, film scanner or the like, is
scanned in an analog-digital converter and each scanned value is linearly
quantized at, for example K=256 levels, to produce a digital signal x
containing an 8 bit word per picture element (pel). A digital prediction
value x, generated by a predictor P, is subtracted from the resulting
digital signal x to form a difference signal e which is quantized in a
quantizer Q which has K' quantizing output levels, where K'<K to produce a
quantized difference signal e.sub.q. This signal e.sub.q is added to x to
form a quantized video signal x.sub.q. It is this signal x.sub.q which
serves as the input signal to predictor P. The quantized signal e.sub.q is
coded in a coder C into constant length code words and is transmitted over
the transmission channel.
At the receiver, the received signal is decoded in a decoder D and the
recovered difference signal e.sub.q is added to prediction value x in
order to regenerate the quantized signal x.sub.q. When the transmission is
perfect, the result of the addition x.sub.q corresponds to the original
value x except for the quantizing error e-e.sub.q, which is the source of
quantization noise. At the same time x.sub.q is fed to a predictor P which
is of identical design to the predictor P in the transmitter and which
regenerates the prediction value x. The regenerated digital signal x.sub.q
is reconverted in a digital/analog converter to an analog signal which can
then be supplied to a monitor.
DPCM methods of this type, even if they employ quantization which utilizes
4 bits per picture element (pel), permitting 16 quantization output
levels, still produce visible quantizing errors, such as "edge busyness"
and "overload", even if complicated two-dimensional predictors are
employed, such as disclosed by D. J. Connor, R. F. W. Pease and W. G.
Scholes in "Television Coding Using Two Dimensional Spatical Prediction",
Bell System Technical Journal, Vol. 50 (1971), at pages 1049-1061.
In order to reduce the visibility of such errors in DPCM systems, methods
have been proposed which employ switchable quantizing characteristics, as
disclosed by H. G. Musmann in German Patent Application No. P 21, 31,
083.8 of June 23rd, 1971.
FIG. 2 shows a block circuit diagram of such a system which can be switched
between n quantizing characteristics. Systems of this type are disclosed
by Th. Kummerow, in "Ein DPCM System mit zweidimensionalem Pradiktor und
gesteuertem Quantisierer" [A DPCM System With Two-Dimensional Predictor
And Controlled Quantizer], NTG-Fachtagung Signalverarbeitung [NTG special
conference on signal processing], April 4th to 6th, 1973 at Erlangen,
Conference Report, pages 425-439. In this system, quantizer Q of FIG. 1 is
replaced by a plurality of quantizers Q.sub.1 . . . Q.sub.i . . . Q.sub.n,
and coder C and decoder D are each replaced by a plurality of coders
C.sub.1 . . . C.sub.i . . . C.sub.n and decoders D.sub.1 . . . D.sub.i . .
. D.sub.n, respectively. Each quantizer corresponds to a respective coder
and decoder. A control logic S at the transmitter decides, on the basis of
the previously transmitted signal value x.sub.q, which quantizer and coder
from the set of quantizers and coders is to be used for each scanned value
to be quantized. An identical logic S at the receiver selects the
corresponding decoder.
The attainable picture quality depends to a great extent on the selection
of the quantizing characteristics and of the control criterion. In
particular, the reduction of quantizing noise which is intended to be
produced with a controlled quantizer can also be adversely influenced when
switching of the quantizing characteristics is effected as a result of
noise in the picture source. These switchings may produce visible
interference if, for example, in a sequence of television images,
different quantizing characteristics are used to quantize a given picture
element of a still picture, and thus the picture element is reproduced
with quantizing noise which changes in time.
SUMMARY OF THE INVENTION
It is therefore an object of the present invention to reduce the
transmission bit rate required to transmit video signals or television
signals of a certain quality.
A further object is to reduce the visible quantizing noise compared to
known methods and to prevent interfering switching effects during
switching between quantizing characteristics.
Another object of the invention is to enable the output bit rate to remain
constant for every picture element.
The present invention, which will be described in detail below, also
utilizes the advantages of switchable quantizing characteristics so as to
reduce quantizing noise. However, the quantizing characteristics and the
control criterion are determined according to a special process and in
that way the subjectively perceivable quantizing noise is reduced compared
to that achieved by the above-noted known solutions and at the same time
the above-mentioned interference effects are avoided.
The difference between the present invention and the technique proposed in
German Application No. P 21, 31, 083.8, cited above, is that in the
latter the absolute value of the prediction value x is used to control the
quantizer while in the present invention the control criterion is derived
in a suitable manner from the quantized difference signal, or prediction
error, e.sub.q.
The difference between the proposal described in the above-cited paper by
Kummerow and the present invention is that in the prior device the
absolute value of the output signal x.sub.q is used to control the
quantizer and the above-mentioned annoying switching effects which occur
in that system are almost completely avoided by the present invention
while the picture quality is substantially increased.
BRIEF DESCRIPTION OF THE DRAWINGS
FIGS. 1 and 2 are block circuit diagrams of prior art quantized signal
transmission systems, which have already been described in detail.
FIGS. 3a and 3b are graphs illustrating the principle of the present
invention.
FIG. 4 is a diagram illustrating relative positions of picture elements in
a scene.
FIG. 5 is a circuit diagram similar to those of FIGS. 1 and 2 illustrating
a preferred embodiment of a system according to the invention.
FIGS. 6a, 6b and 6c are circuit diagrams illustrating construction of a
quantizer according to the invention.
FIG. 7 is a block circuit diagram of an embodiment of control logic for
systems according to the invention.
DESCRIPTION OF THE PREFERRED EMBODIMENTS
A principal novel characteristic of the invention resides in that the
various characteristics are each derived in a suitable manner at K'
quantizing levels from a quantizing characteristic having substantially
more than K' levels. This will be explained with reference to an example
depicted in FIG. 3.
First, the representative values of the quantizing characteristic shown in
FIG. 3b are determined according to the method disclosed by W. Thoma in
"Optimizing the DPCM for Video Signals Using a Model of the Human Visual
System", 1974 Zurich Seminar, Conference volume, pages C3(1) to C3(7) and
by J. C. Candy and R. H. Bosworth in "Methods for Designing Differential
Quantizers Based on Subjective Evaluations of Edge Busyness", Bell System
Technical Journal, Vol. 51 (1972), at pages 1495 to 1516. FIG. 3 relate to
a system composed of a predictor P constructed to effect a planar
prediction for the value of x in the form: x=A+(C-B)/2, the terms of which
are shown in FIG. 4, which depicts a small portion of several lines of a
picture. The current pel is denoted X, while A is the immediately
preceding pel on the same line and B, C and D are adjacent pels on the
immediately preceding scan line of the same field. Pel D is one scan line
less one line element before pel X, pel C is one scan line before pel X
and pel B is one scan line plus one line element before pel X.
FIG. 3a shows the permissible quantizing error for pel X along the ordinate
in dependence on the maximum prediction error e.sub.q at the pels A, B, C
and D along the abscissa. As long as the quantizing error is below this
threshold curve, it will not be visible. The units of both coordinates are
quantum levels of the signal x produced by the analog/digital converter of
FIGS. 1 and 2, where the entire video signal range is divided into 256
quantum levels, level 0 corresponding to black and level +255
corresponding to white. FIG. 3a represents the range of positive quantum
level values. The ordinate of FIG. 3a is in terms of the number of quantum
levels of the quantizing error, e-e.sub.q. The curve was determined by
measurements and indicates that, for example, quantizing errors of less
than 4.5 quantum levels are not visible at element X, i.e. are masked, if
a prediction error of greater than 32 quantum levels has occurred at
picture element A or B or C or D.
Consequently, in this case all representative levels of the quantizing
characteristic forming the basis for the calculations can be eliminated if
their separations are less than 9 quantum levels. Eliminated
representative levels can then be replaced, in order to broaden the
quantizing characteristic for the purpose of reducing the "overload"
effects and "edge busyness", by representative levels added at the outer
ends of the quantizing scale. The principle for selecting and controlling
the quantizing characteristics, which will be explained below with the aid
of an example, is based on these considerations.
In the present invention, each prediction error, e.sub.q, is to be
represented by a 4 bit word, so that no more than K'=16 values, or
quantizer representative levels, can be used.
FIG. 3b illustrates the representative levels, or decision levels, for
several different quantization characteristics. The upper line shows the
representative quantizing levels of the basic quantizing characteristic in
the positive range. The lines identified as characteristics I, II, III and
IV indicate the representative levels of the switched quantizing
characteristics. The locations identified by a "p" exist only in the
positive quantum level range. All other locations designated have a
counterpart in the negative quantum level range (quantum levels 0 to
.+-.256).
The characteristic I of FIG. 3b has the smallest representative levels
concentrated near the 0 quantum levels, and is to be used if the
prediction errors at elements A and B and C and D are all small and thus
only a small quantizing error is permissible.
If, however, the prediction error at any one of picture elements A, B, C or
D is greater than or equal to 20, quantizing errors at point X which are
less than 3.5 will be masked, on the basis of the curve shown in FIG. 3a.
Thus the representative levels 0 and 5 of the characteristic I need not be
used for the quantization of pel X and a representative value 59 can be
added instead to create characteristic II. In a corresponding manner,
characteristic III can be used for pel X if a prediction error greater
than 36 has occurred at any one of picture elements A, B, C or D and
characteristic IV can be used if there are prediction errors greater than
72.
This process can be continued in that further representative levels in the
interior of a quantizing characteristic can be left out and replaced by
higher levels, i.e. higher values of e.sub.q, in such a manner that the
possibly resulting greater quantizing error always remains masked. The
improvement in picture quality compared to an uncontrolled quantizer is a
result of the reduction of the "overload" effects and of "edge busyness"
due to the availability of a greater total number of representative
levels. In comparison with known solutions employing controlled
quantizers, it is here assured that granular noise produced by the small
quantizing stages always remains masked and no visible switching
interference can occur.
FIG. 5 illustrates the general form of a system according to the invention,
which differs from that shown in FIG. 2 in that control logic S, at the
transmitter and at the receiver, is controlled by the quantized difference
signal, or prediction error, e.sub.q.
Table 1, below, sets forth the relationship between the maximum prediction
error of pels A, B, C and D, in number of quantum levels, and the
quantization characteristic of FIG. 3b to be employed for quantizing pel
X. This table also lists the corresponding logic control signal, which
will be discussed below.
Table 1
______________________________________
Magnitude of maximum
Characteristic
Control
prediction error, e.sub.q
to be signal
at pel A, B, C, or D
selected S.sub.1 S.sub.0
______________________________________
.gtoreq. 72 IV O O
.gtoreq. 36 III O L
.gtoreq. 20 II L O
< 20 I L L
______________________________________
One embodiment of such a controlled quantizer for the DPCM system of FIG. 5
is shown in FIG. 6. FIG. 6a shows the subdivision of the quantizer into a
decider E and an evaluator B. The decider E decides into which quantizing
level the prediction error e, represented by 9 bits (e.sub.8 e.sub.7
e.sub.6 e.sub.5 e.sub.4 e.sub.3 e.sub.2 e.sub.1 e.sub.0)belongs. Decider E
emits an 12 bit decision signal E.sub.11 E.sub.10 E.sub.9 E.sub.8 E.sub.7
E.sub.6 E.sub.5 E.sub.4 E.sub.3 E.sub.2 E.sub.1 E.sub.0 and a sign signal
VZ. Table 2, below, lists the relation between the values of prediction
error e and the decisions signals E.sub.11 E.sub.10 E.sub.9 E.sub.8
E.sub.7 E.sub.6 E.sub.5 E.sub.4 E.sub.3 E.sub.2 E.sub.1 E.sub.0.
Table 2
__________________________________________________________________________
Magnitude of the
Decision Signal
Prediction Error, e
E.sub.11
E.sub.10
E.sub.9
E.sub.8
E.sub.7
E.sub.6
E.sub.5
E.sub.4
E.sub.3
E.sub.2
E.sub.1
E.sub.0
__________________________________________________________________________
e = 0 O O O O O O O O O O O O
0 < e < 4
O O O O O O O O O O O L
3 < e < 7
O O O O O O O O O O L L
6 < e < 11
O O O O O O O O O L L L
10 < e < 16
O O O O O O O O L L L L
15 < e < 24
O O O O O O O L L L L L
23 < e < 32
O O O O O O L L L L L L
31 < e < 42
O O O O O L L L L L L L
41 < e < 53
O O O O L L L L L L L L
52 < e < 66
O O O L L L L L L L L L
65 < e < 80
O O L L L L L L L L L L
79 < e < 93
O L L L L L L L L L L L
e > 92 L L L L L L L L L L L L
__________________________________________________________________________
Signals E and VZ are fed to the evaluator B which selects the corresponding
representative levels in dependence on the control signals S.sub.1
S.sub.0, as shown in Table 1, and associates them with a binary number
e.sub.q (e.sub.q8 e.sub.q7 e.sub.q6 e.sub.q5 e.sub.q4 e.sub.q3 e.sub.q2
e.sub.q1 e.sub.q0). The evaluator B can be subdivided into an evaluator
switching network for each digit e.sub.qi (where i=0 . . . 8) of the
binary number e.sub.q, as shown in FIG. 6b.
FIG. 6c illustrates an embodiment of the quantizer section for generating
signal bit e.sub.q0 for the characteristics of FIG. 3b in dependence on
the control signals S.sub.1 S.sub.0. For each characteristic, the value of
bit e.sub.q0 (O or L) is calculated in parallel by logic linkage of signal
bits E.sub.1, E.sub.2, E.sub.3, E.sub.4, E.sub.5, E.sub.6, E.sub.7,
E.sub.9, and VZ, as shown in FIG. 6c. This results in four values at the
outputs of gates 61. The control signals S.sub.1 S.sub.0 switch one of the
four values to the output e.sub.q0. The bits e.sub.q1 to e.sub.q8 are
derived correspondingly.
One preferred embodiment for the control logic S is shown in FIG. 7. A
group of the quantized prediction error bits, e.sub.q4, e.sub.q5,
e.sub.q6, e.sub.q7, is supplied to input AND-gates, some of which have
negated inputs, as shown. The resulting logically linked signals are
conducted along two parallel paths each containing four delay elements
disposed in series and producing, in sequence, delays equal to T.sub.P
(one pel period), T.sub.L -T.sub.P (one scanning line period minus one pel
period), T.sub.P and T.sub.P. The signals at the delay elements are then
logically linked via an array of AND-gates, all but the last of which
having one negated input, to provide bits S.sub.1 and S.sub.0 at the
output of a plural input AND-gate and a plural input OR-gate,
respectively.
The present invention can be used with the same advantage also for other
quantizing characteristics having a different number K' of quantizing
levels and other predictors. Likewise the switching values can be varied
and other combinations of the prediction errors in picture elements A, B,
C, D, which serve as the control criterion, can be used.
It will be understood that the above description of the present invention
is susceptible to various modifications, changes and adaptations, and the
same are intended to be comprehended within the meaning and range of
equivalents of the appended claims.
* * * * *
|
|
 |
|
 |
Description |
|
