IEEE TRANSACTIONS

COMMUNICATIONS,
ON VOL. COM-27, NO. 9 , SEPTEMBER 1979 1335

’ seen fromthe figures, Eq. (15) gives abetterapproximation

than (14) forsmall pf, but is not as good for larger pt.
AdaptiveRoutingRules in ComputerCommunicationNet-
works,” submitted for publication.
[5] T. S. Yum and M. Schwartz, “The Join-Biased-Queue (JBQ) Rule
Finally, we applythisapproximationtechniquetothe anditsApplication to RoutinginComputerCommunication
example of Section 2 . Suppose the following “static” informa- Networks,” submitted for publication.
tion about the network of Fig. l is known: p1 = .9, p2 = .5, [6] G. Foschini and J. Salz, “A Basic Dynamic Routing Problem and
p1 = 6 , p2 = 6.K = 20 for both buffers, Diffusion,” IEEE Trans. Commun., COM-26, 320-327 (1978)
A JSQ decision is t o be made at time t = 4 to routea packet [7] D. R. Cox and W. L. Smith, Queues, Chapman and Hall, London,
196 1
via either node 1 or 2 based on the most recent queue length [8] J. Keilson,“MarkovChainModels-Rarityand Exponentiality,”
information: Report No. CSS 74-01, Center for System Science, University of
Rochester, Rochester, N.Y., Sept. 1974
N,(O) = 0 [9] F. B. Hildebrand, Finite Difference Equations and Simulations,
Prentice-Hall, Englewood Cliffs, N.J., 1968
[ 101 H. Kobayashi, “Application of the Diffusion Approximation to
N2(1) = 10. QueuingNetworks 11: NonequilibriumdistributionsandAppli-
cations to Computer Modeling,”J. ACM21, 459-469 (1974)
Using the first orderapproximation(14) we find from
Fig. 3a and c

fll(4) = 3.9
iTz(4) = 2.5 Coding
Truncation
Block
Image Using
Compression

so that the decision would be to send the packet to node 2 . EDWARD J. DELP, STUDENT MEMBER, IEEE, AND
(The exact values of 3, and 3 2 would yield the same deci- 0 . ROBERT MITCHELL, MEMBER, IEEE
sion.)

5 . CONCLUSIONS Abstract-A new compression

By usinga finite Markovchainmodel,an exact closed
form expression forthetransient behaviorof an M/M/l
queue with finite storage has been derived. It has been shown
that this expression is in a form which lends itself to simple
approximationsforthedynamicbehavior of mean queue
length.Toacertainextent,theresultsobtainedhere
analogous t o those obtained usingdiffusionapproximations.
Specifically, the representation (8) corresponds to the eigen-
function expansion of the solution of the diffusion equation
in
Ref. 10. However, diffusion
approximationsare
only
valid for p =Z 1while ourresultsarenotlimitedtothis
case.
We have alsoillustratedhowtheapproximationscanbe
usedincertainadaptive routing schemes. The fact that the
simple approximations described above have been obtained for
anisolated queue suggests that similarapproachesmaybe
fruitful
for
networks of queues. If reasonably
a simple
dynamicmodelforaqueueingnetworkwerederivedusing
theseapproaches, it wouldbe of considerable value in the
design and analysis of various network control algorithms.
ACKNOWLEDGMENT
The author wishes to thank Prof. D. R. Smith of Columbia
University for calling his attention to Ref. 8.
image
technique
Block
called
for
Truncation Coding (BTC)is presented and compared with transform
andothertechniques.The BTC algorithmusesatwo-level(one-bit)
nonparametric quantizer that adapts t o local properties of the image.
The quantizer that shows greatpromise is one which preserves the local
samplemoments.Thisquantizerproducesgoodqualityimagesthat
appear to be enhanced at data rates of 1.5 bitslpicture element. No
are large data storageis required, and the computation is small. The quantizer
is comparedwithstandard (minimummean-square error and mean
absolute error) one-bit quantizers. Modifications of the basic BTC algo-
rithm are discussed along with the performance of BTC in the presence
of channel errors.
I. INTRODUCTION
Since the beginning of the use of digitaltechniquesfor
image processing, there has been a desire t o find ways of cod-
ing images efficiently. The number of bits required to preserve
high dynamic rangeandresolutionina typical PCM picture
ranges from 1O6 to 10 8 . Many, efficient image coding schemes
have evolved [ 1 1 , popular examples are the Cosine transform
of Chen and Smith [21 and predictive coding (DPCM) [ 31.
In this paper we will present a new technique called Block
Truncation Coding (BTC) that has been developed at Purdue
University in the past two years [ 41 -[ 51 . This technique uses
a one bit nonparametric quantizer adaptive over local regions
of the image. The quantizer that shows great promise is one
that preserves local statistics of the image.l

REFERENCES
Paper approved by the Editor for Data Communication Systems of
[ 1] A. Livne, “Dynamic Routing in Computer Communication Net- the IEEE Communications Society for publication after presentation at
works,”DoctoralDissertation,Dept. of ElectricalEngineering ICC’78, Toronto, Ont., Canada, June 1978. Manuscriptreceived Oc-
and Electrophysics, PINY, 1977 tober 30, 1978; revised March 22, 1979. This work was supported by
[2] C. W. Brown and M. Schwartz,“AdaptiveRouting in Central- the Rome Air Development Center, Griffiss AFB-RADC, under Grant
izedComputerCommunicationNetworks,” h o c . ZCC, San F75C0082 covering the period fromApril 1, 1977 to October 31, 1978.
Francisco, June 1975 The authors are with the School of Electrical Engineering, Purdue
[ 3 ] J. M. McQuillanand D. C.Walden,“TheARPANetwork De- University, West Lafayette, IN 47907.
sign Decisions,” Computer Networks, l 243-289 (1977) At the time of presentation of our ICC paper [ 51 ,a method similar
[4] T. S. Yumand M. Schwartz,“AnalysisofaClassofLocally- to BTC was presented in Japan by Kishimotoet al. [6] -[8].

0090-6778/79/0900-1335$00.75 0 1979 IEEE

1336 IEEE TRANSACTIONS
ON COMMUNICATIONS, VOL. COM-27, NO. 9, SEPTEMBER 1979

In this paper the BTC algorithm is presented and compared

with some common techniques of image compression. Modifi-
cations t o BTC arepresentedincludingsomehybridtech-
niques. The performance of BTC in the presence of channel
errors is also discussed.

11. BASIC BTC ALGORITHM

For the study presented here the image will be divided into
4 X 4 pixel blocks, and the quantizer will have 2 levels. If one
uses the classical quantization design of Max [9] which mini-
mizes the meansquareerror,onemustknow, Q priori, the
probability density function of the pixels in each block. This
same knowledge is also required for the absolute error fidelity
criteria of Kassam [ I O ] . Since in general it is not possible to
find adequate density function models for typical imagery, we
haveused nonparametric quantizers for our coding schemes.
Nonparametricquantizer design that minimizes eithermean
square error (denoted MSE) or mean absolute error (denoted
MAE) will be presented in Section 111. In this section we pre-
sent the designofa nonparametricquantizerthat preserves
sample moments (denoted MP); the design of a parametric MP
quantizer is presented in [ 1 1 1 .
After dividing the picture into n X n blocks ( n = 4 for our
examples), the blocks are coded individually, each into a two
level signal. The levels for each block are chosen such that the Figure 1. Results using the basic BTC algorithm.Theoriginalimages
(top) are 256 X 256 pixels with nominal 8 bits gray level resolution.
first two sample moments are preserved. Let m = n 2 and let The coded images (bottom) have a data rate of 2 bits/pixel.
X , , X,, ... X, be the values of the pixels in a block of the
original picture.
Then the first and second sample moments and the sample Solving for a and 6 :
variance are, respectively

x = -mI r2n xi

(1)

As with the designof anyonebitquantizer, we find a

Eachblock is then described by the valuesof z,
5 andan
n X n bit planeconsistingof 1's and 0's indicatingwhether
threshold, Xth, and two output levels, a and b , such that
pixels are above or below Xth. Assigning 8 bits each to
- 3 and
u resultsinadatarateof 2 bits/pixel.The receiverrecon-
structs the image block by calculating Q and b from Equation 4
and assigningthesevalues t o pixelsin accordancewiththe
code in the bit plane. (An example of coding a 4 x 4 picture
block is presented in the Appendix.)
(2) Examples of this coding system are shown in Figure 1. The
block boundaries are not visible in the reconstructed pictures
For our first quantizer, we set Xth = 3, thisreasonable as- nor is the quantization noise visible in regions of the picture
sumption will bemodifiedinSectionIV to improveper- where there is little change in luminance. Here the levels Q and
formance. The output levels Q and b are found by solving the 1, are close together. They are widely spaced only in regions
following equations: where large changes of luminance occur; but it is well known
[ 131 that large changes of luminance mask noise. Thus BTC
Let q = number of Xi's greater than Xth(=x) encoding makes use of the masking property in human vision.
- The most noticeable improvementof the reconstructed picture
then to preserve x and X 2 is a little raggedness of sharp edges. In the next section we will
present methods for improving the image quality and reducing
m y = ( m -4)" + qb (3 1 the bit rate.
Because the calculations are relatively simple and the stor-
and age of data small, BTC is fairly easy to implement. A recent
study has shown that BTC can be realized on an integratedcir-
cuit chip [ 121.
TRANSACTIONS
IEEE ON COMMUNICATIONS, VOL. COM-27, NO. 9 , SEPTEMBER 1979 1337

111. OTHER NONPARAMETRIC QUANTIZER SCHEMES

As mentioned in Section 11, other techniques can beused to
design a one bit quantizer. Use of rate-distortion theory seems
theoreticallyattractivebutsomewhatimpracticalforreal
images [ 141-[ 151. In this section we will discuss the use of
the minimum mean square error (MSE)andminimummean
absoluteerror(MAE)fidelitycriteria for one-bitnonpara-
metric quantizers.
To use the MSE fidelitycriterion,oneproceedsbyfirst
constructingahistogramofthe Xi’s (i.e., sorting the Xi’s).
< <
Let Y , , Y 2 , ... Y m be the sorted Xi’s; i.e., Y1 Y z ... Y m .
Again let q be the number of Xi’s greater than Xth. Then a and
b are found by minimizing

i=l i=m-q

where

Figure 2. Results using various fidelity criterion. All data rates are 2.0
l m bits/pixel.Upper left:minimummeansquareerror; Upperright:
b=- C Yi minimummeanabsoluteerror;
Lower right: minimum mean
Lowerleft: moment preserving;
square error and also assuming image
4 i=m-q
data uniformly distributed each block.

One obvious way to solve this problem is to try every possible

threshold (there are at most m - 1 thresholds) and pickthe one
with smallest JMSE. Assuming Q and b have 8-bit resolution,
this gives a data rate of 2 bits/pixel.
Theproblem o f using the MAE fidelitycriterion is very
similar to the MSE. The values Q and b are found by minimiz-
ing

where

Q = median of (Y1, Yz, ..., Ym-q-1)

b = median of ( Y m W q ,--, Y m ) .

Hereagain thenonparametricquantizer is arrived at by an

exhaustive search. Results using these quantizers and BTC are
shown in Figures 2 and 3. As mentioned in Section I1 it is
possible to useaparametricquantizeronce theprobability
densityfunctionofthe pixels is known(or guessed).In
Figures 2 and 3 resultsarepresentedforaparametric MSE
quantizerwherethe pixels are assumed t o beuniformly Figure 3. These four pictures were produced as described in Figure 2
distributed over each block. Table I has the computed mean using the other original.
square error and mean absolute error measures for each image.
As anticipated the MSE quantizer has the smallest computed
meansquareerrormeasureand the MAE quantizer has the IV. BTC MODIFICATIONS
smallest computedmeanabsoluteerrormeasure.The per- In many image coding schemes it is desired to obtain data
formance ofBTC is quitegoodwhencomparedtothese rates less than 2 bits/pixel. An obvious way of achieving this
standard fidelity criteria. The advantage of using a nonpara- is t o assign less than 8 bits to and 5.Experimental evidence
metric MP quantizer is thatthequantizerformulation is has indicated that coding 7 with 6 bits and a
with 4 bits in-
available in closed f o r m ; this greatly simplifies the computa- troduces only a few perceivable errors. Now the data rate is
tional load. 1.63 bits/pixel. Better performance could be obtained byw a n -
 1338 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. COM-27, NO. 9, SEPTEMBER 1979

TABLE I
COMPUTED MEAN SQUARE ERROR AND MEAN ABSOLUTE
ERROR MEASURES FOR VARIOUS QUANTIZATION
SCHEMES
Mean Square E r r o r Mean Absolute
Error
b a t a from Figure 2:
Using MSE q u a n t i z e r 3.54 32.94
Using MAE q u a n t i z e r 37.13 3.28
Using BTC 40.89 3.91
Using
parametric MSE q u a n t i z e r 44.64 4.23
and assumed uniformdensity

Datafrom Figure 3:
Using MSE q u a n t i z e r 4.39 47.14
4.10 MAE q u a n t i z e r
Using 53.22
Using BTC 58.34 4.85
Using
parametric' MSE q u a n t i z e r 64.02 5.42
and assumed uniformdensity

a
tizing 2 and jointly, using 10 bits. f is assigned most bits in
blocks where 0 is small and fewest bits where 0 is large.
Additional savingscouldberealizedbyefficientlycoding
the bitplane.Entropycalculationsindicatethat 0.85 bits/
pixelwouldbesufficientfor the bitplane.Suchsmallim-
provementdoes not justify the increased complexity of the
system and the increased variability to error that efficient cod-
ing entails.
Setting the threshold of the quantizer 2, at removes a possi-
ble degree of freedom for optimizing the encoding. Allowing Figure 4. Results of BTC using third moment preserving threshold
the threshold tobea variable permitstheencodingtopre- selection. Original images are atthetop.Codedimages are atthe
serve not on1.y the first two sample moments but also the thud bottom. Data rate is 2.0 bits/pixel.
sample moment. To analyze this, we make use of the sorted
pixel-values Yi. The third moment can beexpressed as In general the value of q obtained will not be an integer,so
inpracticeitmust be roundedtothenearestinteger. An
interesting interpretation of Equation 9 is that the threshold
is nominallyset to the samplemedianandbiasedhigher or
lower depending upon the value of the third sample moment
The problem then is finding a, b, and q to preserve 2,f 2 , and which is ameasureofskewnessof the Yi's. Thisthreshold
' T3 ( q definesthethresholdsince it specifies thenumberof selectionrequires no extra computation at the receiver(de-
Xi's greater than X , , ) , Equation 3 now becomes coder), but the transmitter (encoder) does have extra process-
ing. This method of threshold selection is far easier to imple-
m X = ( m - 4)" + qb ment than are the nonparametric quantizers discussed in Sec-
tion I11 since they require an exhaustive search.
Figure 4 shows results using this threshold selection. It im-
proves subtle features (such as near edges) of the image that
are usually important in analysis of aerial photography imagery.
This improvement will be discussed in the following section.
Having the solutions
V. PERFORMANCE EVALUATION OF BTC
Recent studies at Purdue University have evaluated various
image coding techniques for transmitting aerial reconnaissance
imagesovernoisychannels [ 161. BTC, transform,Hybrid
[ 17] and one other technique were evaluated by professional
photo analysts.
Although BTC did not perform as well as transform coding
in the photo interpreters evaluation, it proved superior to the
other techniques. In the presence of many channel errors, BTC
was superior to all of the other techniques. Some images used
in the study are shown in Figures 5-8. In these figures BTC is
compared with the Chen and Smith [ 2 ] method of transform
where
coding and Hybrid coding. The computed mean square errors
and mean absolute errors are shown in Table 11. Our study has
shown that the meansquareerrorandmeanabsoluteerror
measurecannot beeasilycorrelatedwith photo analysts'
evaluations.Insome cases, imageswithlargemean square
-
u # 0. errors wereevaluatedhigherthan imageswithsmallermean
square errors. It should be noted that BTC requires a signifi-
Ifa=O,thena=b=X. cantly smaller computational load and much less memory than
IEEE TRANSACTIONS
ON COMMUNICATIONS, VOL. CUM-27,NO. 9 , SEPTEMBER 1979 1339

Figure 7. Results of coding original (Figure 5) using Chen and Smith

[2] method of Cosine Transform coding. Data rate is 1.63 bits/pixel.

Figure 5 . Originalimage usedin comparison study. Image is 512 X

5 12 pixels with nominal 8 bits gray level resolution.

Figure 8. Resultsofcodingoriginal(Figure 5) using Hybridcoding

[17]. Data rate is 1.63 bits/pixel.

TABLE I1
COMPUTED MEAN SQUARE ERROR AND MEAN ABSOLUTE
Figure 6 . Results of codingoriginal(Figure 5 ) using BTC with third ERROR MEASURES FOR COMPARISON IMAGES SHOWN IN
moment preserving threshold selection. Data rate is 1.63 bits/pixel. FIGURES 6-1 1 AND FIGURE 13
Mean Square E r r o r Mean Absolute E r r o r '
Figure 6 84.22 5.94
Figure 7 67.15 6.32
Figure 8 125.84 6.12
Figure 9 115.09 6.29
Figure10 115.31 7.06
Figure 11 140.33 6.67
Fisure13 74.67 5.72
1340 IEEE TRANSACTIONS
ON COMMUNICATIONS, VOL. COM-27, NO. 9 , SEPTEMBER 1979

Figure 9. BTC coding withchannelerrorprobability of 10-3. Data Figure 11. Hybridcodingwith channelerrorprobability of 10-3.
rate is 1.63 bits/pixel. Data rate is 1.63 bits/pixel.

Figure 12. Enlargedsection of image from Figure 6 (left) and Figure

7 (right).

channel was assumed t o be binary symmetric with the proba-

bility of a bit error of 1O - 3 .
One of the advantages of BTC is that luminance edges are
emphasized. Figure 12 shows an enlarged section of an image
coded using the Chen and Smith method and BTC. The edges
using the transform coding method are not as sharp as the BTC
image.
As with all noninformation preserving image coding, coding
Figure 10.ChenandSmith coding withchannelerrorprobabilityof artifacts are produced in the image. It became apparent very
10-3. Data rate is 1.63 bits/pixel.
early in this study that BTCprodu,Ges artifacts that.are very
differentthantransform coding. Theyare usuallyseen in
transform or Hybrid coding. For instance, the Chen and Smith regions around edges and inlowcontrast areascontaininga
methodrequiresthetwo-dimensional Cosine transform over slopinggray level. As mentionedabove, BTC doesproduce
every 16 X 16 imageblockandalsorequiresmultiple passes sharp edges;however,theseedges do havea tendency to be
through the transform data to collect various statistics about ragged. Transformcodingusuallyproducesedgesthatare
the transform coefficients. Itshould also be mentioned that blurred and smooth. The second problem in low contrast re-
BTC requires no sophisticated error protection as do the other gions is duetoinherentquantization noisein theonebit
codingmethodsevaluated. Figures 9-1 1showtheresults of quantizer. Here sloping gray leveis can turn into false contours.
eachcodingmethodinthe presence of channel errors. The Preliminaryexperiments haveindicatedthatpre-andpost-
TRANSACTIONS
IEEEON COM~UNICATIONS,VOL. COM-27, NO. 9 , SEPTEMBER 1979 1341

structed image hasartifactsquitedifferentfromthose pro-

duced by other techniques, but the resulting images are com-
parable and in some ways superior to those produced by the
most sophisticated techniques available. This method produces
a coded image that is more robust in the presence of channel
errors and requires very little error protection overhead.

ACKNOWLEDGMENT
The authors would like to thank Dr. John Gamble of the
Rome Air Development Center for his helpful criticism of this
work. We would also like to thank Richard P. Petroski and his
colleagues of the Rome Research Corporation whoprovided
the expert photoevaluations mentioned in Section V.
The authors would also like to thank Professor Steven Bass
of Purdue for suggesting the hybrid formulation of BTC and
Ali Tabatabai and Paul Stiling also of Purdue for providing the
transform and Hybrid coded images.

Figure 13. Results of HybridFormulation of BTC: Data rateis 1.88
bits/pixel.
processing of the image can reduce the effects of both these
artifacts while simultaneously reducing the mean square error
and mean absolute error. It should be emphasized that these
coding artifacts are problems in high resolution aefial recon-
naissance images where minute objects are important. These
coding artifacts usually are not visible intypical “head and
shoulders” imagery.
APPENDIX
Toillustrate coding a4 X 4picture block using Block
Truncation Coding, let usquickly review the basic BTC
algorithm:
a) the image is divided ih small nonoverlapping blocks such
as 4 X 4.
b) The first andsecond sample moments are computed.
c) A bit plane is constructed such that each pixel location is
coded as a “one” or a “zero” depending on whether that pixel
is greater than x.
d) The bitplane, $, and 0 are sent to thereceiver.
x
e) The picture block is reconstructed such that and (al-a
ternatively X 2 ) are preserved. That is, pixels in the bit plane
that afe “0” are set to “a” and the “1”’s are set to “b” in
Equation 4. For example, suppose a 4 X 4 picture blockis
given by the following:

[ :z
VI. HYBRID FORMULATION OF BTC
121 114 56 47
‘BTC uses only first-order informationanddoesnot ex-
ploit the two-dimensional sturcture of the image within each 37 200 247 255
block as do most other forms of image coding. For example xij= 12 16g]
in two-dimensional transform coding thetransform coeffi-
cients contain information about variations in the picture in 251
7
both directions. Also BTC generally has a poor response near
the spatial frequency of 1/2 cycle per block. so
One methodto overcomethisdisadvantage is ahybrid
formulation. First a highly compressed Cosine transform
x = 98.75
coded image is subtracted from the original image. For the re- -
u = 92.95
sults presented herethetransformpicture was obtained by
taking the two-dimensional Cosine transform over 16 X 16 q=7
pixel blocks. Only the eightnon-dc coefficients in thelow
frequency section of eachblock were retained. Thiscor- and
responds to a zonalfiltering method with abitrateof
0.25 bits/pixelforthe highlycompressed image. BTC was a= 1 6 . 7 s 17
then used on this difference pictureandtherecombination
formed at the receiver. While this did increase the computa- b = 204.2 S 204
tional load, the improvement seemed to be significant enough
the bit planeis
to give this method further attention. Figure 13 presents re-
sults of this hybrid method. TableI1 has the mean square error
and mean absolute error measures for Figure 13.

VII. CONCLUSIONS
A new,simple technique has beendescribedwithmodifi-
cationsto improve itsperformance.The resulting
recon- Lo o o IJ
The reconstructed block becomes Effects of Digital Demodulation on Component Coding o f
r 2 0 417 17
204 NTSC Color Signals
204
204
17
204 ERIC DUBOIS

Abstract-Effects of nonideal filtering in the digital demodulation

of the composite NTSC signal into luminance and chrominance com-
and the sample mean and variance are preserved. ponents on the coding of the luminance component are discussed. The
residualchrominancecomponent is shown t o causedegradationin
REFERENCES prediction and/or motion detection algorithms which are based o n line
or framedifference signals. Simple techniques for reducing these effects
A. Habibi,“SurveyofAdaptiveImageCodingTechniques,” using chromiqance-band rejection filters are presented and the efficacy
IEEE Trans. on Communications, Vol.COM-25,No. 11,pp.
of these methods areverified by simulation.
1275-1284, NOV. 1977.
W-H. Chen and C. H. Smith, “Adaptive Coding of Monochrome
andColor Images,” IEEE Trans. on Communications, Vol.
COM-25, pp. 1285-1292, NOV.1977. I. INTRODUCTION
J. B. O’Neal,“PredictiveQuantizingSystem (DPCM) forthe
TransmissionofTelevisionSignals,” BSTJ, Vol. 45,pp.689- Acommontechniqueforcoding NTSC colorsignals for
721, May-June 1966. bandwidthcompression is component coding,whereby the
0. R. Mitchell and E. J. Delp, “Image Compression Using Block composite NTSC signal is separated intoluminanceand
Truncation,”Purdue University-F’urdueResearch Foundation,
Record and Disclosure of Invention, dated April 8, 1977. This chrominancecomponents,andeach is codedindividually.
document i s available from the authors. There are several advantages to performing this demodulation
0. R. Mitchell, E. J. Delp, and S. G. Carlton, “Block Truncation using digital filtering techniques [ 11. Analog demodulators re-
Coding: A New Approachto ImageCompression,” Conference quire the use of high precision analog delay lines which may
Record, 1978 IEEE International Conference on Communica- require frequent alignment, while, of course, digital filters re-
tions (ICC’78), Vol. I, June 4-,7, 1978, pp. 12B.l.l-12B.1.4.
T. Kishimoto, E. Mitsuya, and K. Hoshida, “A Method of Still quire no such alignment. Also, in all-digital installations, the
PictureCoding byUsing StatisticProperties,” (inJapanese) use of analogfilterswouldrequireadditionalD/AandA/D
National Conference of the Institute of Electronics and Commun-conversion.However, due to the high dataratesassociated
ications Engineers o f Japan, March 1978, No. 974. with color video signals, only relativelylow order filters are
T. Kishimoto, E. Mitsuya, and K. Hoshida, “An Experiment of
Still Picture Coding by Block Processing” (in Japanese) National currently cost effective. Digital demodulation using such filters
Conference of the Institute of Electronics and Communications has not been widely used in color codecs, partly due to the
Engineers of Japan, March 1978, No. 975. residual chrominance effects which will bedescribedinthis
T. Kishimoto,E.Mitsuya,and K. Hoshida,“BlockCoding of paper.
Still Pictures,” (in Japanese), Technical Group of Communica- The frequency of the color subcarrier in an NTSC signal is
tion System of the Institute of Electronics and Communication
Engineers of Japan, pp. 63-69, July 1978. such that it changes in phaseby 180’ from line to line and
J. Max, “Quantlzing’for Minimu’m Distortion,” IRE Trans. on fromframetoframe. When thecomposite NTSCsignal is
Info. Theory, Vol. IT-6, pp. 7-12, March 1960. separated intoluminanceandchrominancecomponentsby
S . A. Kassam, “Quantization Based on the Mean-Absolute-Error digital filtering, a residual chrominance component remains in
Criterion,” IEEE Trans. on Communications, Vol. COM-26, pp.
theluminancecomponentduetononidealfiltering.Taking
267-270, Feb. 1978.
E. J. Delp and 0. R. Mitchell, “Some Aspects of Moment Pre- a frame difference or line difference of this luminance signal
servingQuantizers,”in Con$ Rec., I979 IEEE International causes the residual chrominance component t o be reinforced
Conference ,012 Communications, (ICC’79),vol. I,June10-14, due to the 180‘ phase shift of the subcamer, while the lumi-
pp. 7.2.1-7.2.5. nancecomponent cancels to alarge extent,duetoline-to-
W. L. Eversole, D. J. Mayer, F. B. Frazee, an T. F. Cheek, “In-
vestigationofVLSITechnologies for ImageProcessing,” Pro- line or frame-to-frame correlation. This results in a substantial
ceedings: Image Understanding Workshop, Pittsburgh, PA, Nov. spuriouscomponentinthedifference signal, locatedinthe
14-15, 1978. Sponsored by the Defense Advanced Research Pro- chrominance band, causing degradation in the performance of
jects Agency (DARPA), pp. 191-195. (Copies available from the prediction and/or motion detection algorithms based on line
DefenseDocumentationCenter(DDC)underAccession No.
ADA 052903).
orframedifferences.Theseeffectscanbeminimizedby
D. Jameson and L. M. Hurvich (editors), Handbook o f Sensory passing thedifference signal through a chrominance-band
Physiology: Visual Psychophysics, New York;Springer-Verlag, rejection filter, similar to that used t o separate the luminance
1972. fromthecomposite waveform. This
paper
examines the
T.Berger, Rate Distortion Theory, EnglewoodCliffs:Prentice- application of this principle to a number of schemes for coding
Hall, 1971.
J. L. Munnosand D. J. Sakrison,“TheEffects of a Visual the luminance componentof the NTSC signal.
Fidelity Criterion on the Encoding of Images,’’ IEEE Trans. on
Information Theory, Vol. IT-20, pp. 525-536, July 1974.
0. R . Mitchell, S. C. Bass, E. J. Delp, and T. W. Goeddel, “Coding Paper approved by the Editor for Data Communication Systems of
ofAerialReconnaissanceImagesforTransmissionoverNoisy the IEEE Communications Society for publication after presentation at
Channels,”issuedasRome Air DevelopmentCenter(RADC) the 9th Biennial Symposium on Communications, Queen’s University,
TechnicalReport, Griffiss Air Force Base, NY. Report No. Kingston, Ont.,Canada, August 1978. Manuscript received September8,
RADC-TR-78-210.(This report is availablefrom theNational 1978; revised April 2, 1979. This work was supported in part by the
Technical Information Service, Accession No. ADA 061539). National Research Council of Canada underGrant A0022.
A.Habibi,“HybridCoding of PictorialData,” IEEE Trans. The
author is with INRS-Telecommunications, University of
Comyun., Vol. COM-22, pp. 614-624, May 1914. Quebec, Verdun, P.Q., Canada H3E 1H6.

0090-6778/79/0900-1342 $00.75 0 1979 IEEE