International Journal of Engineering Trends and Technology (IJETT) – Volume 34 Number 7- April 2016
Print quality assessment for web offset colour
printing machine in Newspaper production
Srividya B1, Thirunavukkarasu V2
1
Research Scholar, Anna University,
Assistant Professor, Department of Printing Technology, Avinashilingam University, Tamil Nadu, India.
2
Professor, Dept. of Mechanical Engineering, EGS Pillay Engineering College, Tamil Nadu, India.
1
Abstract - Job and production processing steps in print
production are highly automated, except print quality
assessment. Forests-based technique is used for
automatic print quality assessment in newspaper
production based on independent values of several
attributes based on print quality. Recent developments
for assessing the print quality are image analysis and
computer intelligence based techniques since manual
analysis are time consuming and results are dependent
on observer. The soft sensors are used to determine the
values for attributes through data mining and image
analysis (only colour). The experimental shows the better
result for print quality evaluations. The total of 25 print
quality attributes are measured automatically by testing
the double grey bars on the newspaper and it can be used
for expert system. Although mapping, random forest and
clustering techniques have been used for investigating
the print quality. Report shows that even min number of
attributes can score max accuracy in prediction of print
quality.
Keywords: Print quality, Print quality assessment,
Random forest technique, double grey bar, web
offset printing, Newspaper production.
I. INTRODUCTION
Offset printing is the most widely used
commercial printing process in production of
newspapers and magazines [1]. A multicolour
image comprised of cyan, magenta, yellow and
black dots of varying size usually called as halftone
image. These halftone dots are then transferred to
the printing plate through computer to plate
process. Each colour takes up the separate plate for
production. During production of newspaper
various factors that affects the print quality are
paper roughness, ink density, printing press, ink
water balance, plate quality and processing and
operator efficiency. Major deficiencies and
problems occur only during printing; paper surface
(wrinkle, smoothness, linting, mottling and web
breaks) and mis-registration of plates used for
colour printing. Even though these problems are
overcome still the quality of the print can be low
since the assessment on print quality attributes are
only through manual. Also values for print quality
attributes are altered timely for instant requirement
to improve the quality. Fig.1 shows the double grey
bar which is printed on the edge of the newspaper
ISSN: 2231-5381
used for visual and instrumental assessment of print
quality attributes
Fig.1 Double grey bar
This paper is concern with random forest based
technique for assessing the print quality of the web
offset printing machine the values are estimated
automatically on double grey bar image
II. RECENT PROGRESSES IN PRINT
QUALITY ASSESSMENT
Recently image analysis and computer
intelligence based techniques are used for assessing
the print qualities since manual procedures are time
consuming and results are dependent on observer.
Developments in this area usually concern
simulation of the most common print quality
defects [3,4], inspection of one or several print
quality attributes, such as the actual size and
quality of printed dots [5], mottling [6], CCD
camera-based estimation of ink density [7],
automatic detection and classification of various
printing defects [8]. A recent study used Bayesian
networks and genetic algorithms to model the
overall print quality assessment by a group of
people in electrophotography printing [9]. Guan et
al. have recently developed a case based reasoning
system for offset print quality control [10]. A
variety of print quality cases are stored in a
knowledge base and exploited for decision-making
in the printing process. [2]. Perner developed a
knowledge based image inspection system for
automatic defect detection, classification, and
misprint diagnosis in offset printing [8]. A CCD
linear camera is used as an image sensor. The
system can recognize defects, categorize them into
one of 47 classes, including colour drift, and
suggest actions for the operator to eliminate the
cause of defect. Another image analysis based
defect detection tool, assisting the press operator in
http://www.ijettjournal.org
Page 322
International Journal of Engineering Trends and Technology (IJETT) – Volume 34 Number 7- April 2016
finding defect in the print and taking appropriate
adjustments, was tested on a Flexo gravure printing
press [11, 12]. The defect detection is based on a
threshold difference image between the reference
print and the sample print. A similar approach of
comparing the reference print and the sample print
is also utilized in [13]. A print quality inspection
system developed by Brown et al. for a rotogravure
printing press wastested in the wall-covering
printing industry [14, 15]. Chuang and Lai [16]
developed a printing quality control expert system
for controlling offset printing dot variations. The
system consists of knowledge base, the inference
engine, and the user interface. Asikainen [17]
developed a linear model to predict a value of an
overall image quality index using four quality
attributes, namely colourfulness, contrast, noise,
and colour difference, measured on a reference
image taken from a picture printed with an ink jet
printer. Good correspondence between the model
and human assessments of the overall quality was
found [2].
III. PRINT QUALITY ATTRIBUTES
The various print quality attributes are listed
here are used in this work:
a.
b.
1,2: Dot Formation (sd, fs )
3,4: Dot gain (tonal value on black area of
double grey bar & standard deviation)
c. 5-10: Mis- registration ( cyan, magenta
and yellow in both X and Y direction)
d. 11-14: Colour deviation (∆E of L*a*b*)
e. 15,16: Contrast (black and colour grey
bar)
f. 17-20: Noise level (black, cyan, magenta
and yellow ink in a grey bar)
g. 21: Ink deviation (from actual level)
h. 22-25: Ink density (black, cyan, magenta
and yellow)
Altogether 25 parameters are used for assessing the
quality of the print on double grey bar image.
These attributes are evaluated by capturing the
image on the digital image camera (CCD) for
interpretation of values.
IV. SUBJECTIVE ASSESSMENT
PROCEDURE
The experts (having knowledge of printing
& quality) group of two and non-expert groups of
five are teamed for evaluating print quality. For
investigation MHI Diamond Spirit of sixcolour web
offset printing machine is included for testing on
the newspaper. This paper concentrates on two
ISSN: 2231-5381
experiments overall quality of print and
comparative analysis of various defective prints
using 10 point scale, where 10 indicates highest
quality or similarity between the prints. The first
test uses the florescent lamp of colour temperature
3000 K [2] and second test was conducted in shop
floor where printing was done with lamp colour
temperature of 6500 K.
V. MONITORING THE PRINT QUALITY BY
MACHINE
Temponi et al. suggested using grey
balance, tone reproduction, and overall appearance
as printing quality attributes [18]. To assess the
overall printing quality level, the quality attributes
are combined as a triangular fuzzy number [2]. In
[19],it was suggested using fuzzy integration to
aggregate print quality attributes, into an overall
print quality measure. A system attempting to
simulate human print quality assessment for simple
prints made by laser and ink jet printers was
presented in [20]. Using simple print features
characterizing noise level, edge sharpness, and
tonal contrast the system was trained to categorize
prints into the ”bad quality” and ”good quality”
classes. Both linear and nonlinear models were
used to predict a print quality score by machine
using print quality attributes as independent
variables in this work. Random forest (RF) [21]
was the main nonlinear modelling technique
applied. Random forests have been successfully
applied in a variety of fields and have several
appealing properties. Due to low computational
complexity, RF can handle thousands of variables
of different types with many missing values [2].
A. Random Forest (RF)
Random forest, a general data mining tool
proposed by Breiman [21], is a committee of weak
learners—decision trees, i.e. CART [22]. A weak
learner can be created using a training setX. A
weak learner is a predictor f(x,X) having low bias
and high variance. By randomly sampling from the
set X, a collection of weak learners f(x,X, θk) can
be created, where f(x,X, θk) is the kth weak learner
and θk is the random vector selecting data points
for the kth weak learner [2]. When solving
classification problems, RF prediction is the unweighted majority of class votes. As the number of
trees in RF increases, the test set error rates
converge to a limit, meaning that there is no overfitting in large RFs [21]. For accuracy low bias and
low correlation are essential.
http://www.ijettjournal.org
Page 323
International Journal of Engineering Trends and Technology (IJETT) – Volume 34 Number 7- April 2016
VI. INVESTIGATION PROCESS
A. Subjective assessment of print quality
Test samples of thirty printed on 45 GSM
News print Paper (gloss) coated grade with long
grain of size 505 x 735 mm. The print quality was
evaluated for registration defects, ink deviation,
noise due to ink water unbalance and it’s indicated
in the Fig.2.and related standard deviation indicated
in Fig.3.
Fig.4. GATF Test form printed
Fig.2. Average quality prints of subjective
C. Evaluation of subjective print quality
Evaluation of subjective print quality
required random forests, mapping and clustering
technique. Based on the first experiment mapping
was done to evaluate the subjective assessment
score and he resultant data was further explore by
random forests. The important task is to determine
the most suitable print quality attributes for
determining the principle directions of the mapped
data. From the Fig.5 it is found that 3 misregistration of magenta printing plate (X direction),
4 mis-registration of yellow printing plate (Y
direction), 12 L*a*b component values and 4
contract values are assed in the double grey bar
image area considered as most important factors for
assessment.
Fig.3. Standard deviation of average quality prints
of subjective
B. Print quality attributes measurement by
machine
Attributes related to print quality were
measured on the double grey bar printed on the end
of the newsprint paper. Fig.1 represents the
example of double grey bar which is measured on
X and Y position of cyan, magenta and yellow
parts of the bar with respect to the black. It is used
to detect the mis-registration, colour deviation
noise level, dot formation and dot again.Even
though ink density can be estimated on grey bars
using the technique suggested in [7],measurement
of ink density by densitometer on a solid print areas
of the test newsprint prints, see Fig. 4. A data set is
collected to implement the technique suggested in
[7].
ISSN: 2231-5381
Fig.5.Print quality attributes for assessment
D.Predicting print quality score
By considering the most important factor
from subjective evaluation random forests tree was
created and selective variables are sliced according
to the 25 print quality attributes. Average values of
the subjective evaluations were used as the source
data for designing the model. The three models
simulated average subject, average expert, and
average non-expert. First, the three models were
designed using all 25 quality attributes as input
variables as mention in the [2]. Also it was found
that high score of print quality was obtained only
when using four attributes 7,10,13,and 16 it is
shown in the Fig.6. The average non-expert model
provides more accuracy than average subject for
the attributes 7, 10, 13 and 16. Also it is easy to
http://www.ijettjournal.org
Page 324
International Journal of Engineering Trends and Technology (IJETT) – Volume 34 Number 7- April 2016
distinguish two maps either by lower or higher than
average value of print quality. Linear regression
model with d= 1, 2, 3 was created using all 23
attributes for comparing the random forests score.
when we use limited number of attributes for
assessment because of large variety of papers, inks
and control parameter. Also there is a symptom of
accuracy when experts assessing the print quality
than non-expert, for mis-registration and colour
deviation. The choice of the user, alternative
equipment and materials are the factors for
affecting the product quality.
ACKNOWLEDGEMENT
Authors wishes to acknowledge J. Lundstrom A.
Verikasfor developing the procedure for
assessment of print quality by subjective observer
Fig.6. Print quality attributes for average subject
model
E. Print quality comparisons for pair wise
Pair wise comparison was done for second
experiment using 30 print samples for evaluating
the most important variables like mis-registration
and lab values. Clustering analysis were carried out
for the above two variables since it occupies the
major area of the print in the print sample it is
showed in the Fig.7. Finally the Lab component has
the high score and having negative impact in the
test result.
REFERENCE
[1]
H. Kipphan (Ed.),(2004)Handbook of Print Media,
Springer.
[2]
J. Lundstrom A. Verikas, (2012)Assessing print quality by
machine in offset colour printing, Knowledge-Based
Systems, Elsevier, 0950-7051
G. E. Kowalczyk, R. M. Trksak, (1998) Image analysis
of ink-jet quality for multi use office paper, TAPPI
Journal 81 (10) 181–190.
W. Jang, J. P. Allebach, (2005) Simulation of print quality
defects,Journal of Imaging Science and Technology 49
(1) 1–18.
C. Antoine, M. D. Lloyd, J. Antoine, (2001) A robust
thresholdingalgorithm for halftone dots, Journal of Pulp
and Paper Science 27 (8) 268–272.
A. Sadovnikov, L. Lensu, H.Kalviainen, (2007)
Automated mottling assessment of colored printed areas,
in: Image Analysis, Proceedings, Vol. 4522 of Lecture
Notes in Computer Science, Aalborg, Denmark, , pp.
621–630.
A. Verikas, M. Bacauskiene, (2008) Estimating ink
density from colour camera RGB values by the local
kernel ridge regression, Engineering Applications of
Artificial Intelligence 21 (1) 35–42.
P. Perner, (1994) Knowledge-based image inspection
system for automatic defect recognition, classification
and process diagnosis, Machine Vision and
Applications7 135–147.
T. Eerola, L. Lensu, J.-K. Kamarainen, T. Leisti, R.
Ritala, G. Nyman, H. Klviinen, (2011) Bayesian network
model of overall print quality: Construction and
structural optimisation, Pattern Recognition Letters 32
1558–1566.
L. M. Guan, J. Lin, G. J. Chen, M. Chen, (2006) Study for
the offset printing quality control expert system based on
case reasoning, in: Proceedings of the 2006
IEEE/ASMEInternational Conference on Mechatronic
and Embedded Systems and Applications, IEEE/ASME,
Beijing, P.R.C, pp. 272–276.
J. L. Shankar, N. Ravi, Z. W. Zhong, (2003) On-line
defect detection in web offset printing, in: Proceedings of
the 4th International Conference on Control and
Automation,IEEE, Montreal, Canada, pp. 794–798.
J. L. Shankar, N. Ravi, Z. W. Zhong, (2009) A real-time
printdefectdetection system for web offset printing,
Measurement 42 645–652.
H. Kipphan, W. Geissler, G. Fischer, W. Huber, B.
Kistler, G. Bucher, C. Rensch, Process and arrangement
for controlling or regulating operations carried outby a
printing machine, U.S. Patent 6050192, filed April 29,
1996, and issued April 18, 2000.
[3]
[4]
[5]
[6]
[7]
[8]
[9]
Fig.7. Test sample for measuring print quality
attributes
[10]
VII. CONCLUSION AND FUTURE WORK
The effectiveness of the automatic print
quality assessing system depends on how well the
result correlates with human judgements. An
automatic print quality assessment system used in
this work for measuring on the double grey bar
image that can also be used in expert support
system in paper making and printing industries.
Even though there is a disagreement between
subjective assessmentsand
prediction from
designed model are slightly encouraging. The
effectiveness of the system can be determined only
ISSN: 2231-5381
[11]
[12]
[13]
http://www.ijettjournal.org
Page 325
International Journal of Engineering Trends and Technology (IJETT) – Volume 34 Number 7- April 2016
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
N. Brown, M. R. Jacksson, P. E. Bamforth, (2004)
Machine vision in conjunction with a knowledge-based
system for semi-automatic control of a gravure printing
process, Proceedings of the Institution of Mechanical
Engineers, Part I: Journal of Systems and Control
Engineering 218 (4) 583–593.
N. Brown, M. R. Jacksson, R. M. Parkin, (2003)
Automatic gravure print feature determination at
production speeds, Proceedings of the Institution of
Mechanical Engineers, Part B: Journal of Engineering
Manufacture 217 (8) 1101–1110.
C. P. Chuang, F. P. Lai, (1997) Developing a prototype of
qualitycontrol expert system for offset printing dot
variations troubleshooting, in: TAGA 1997: Proceedings
– Disseminating Graphic Arts Research Internationally
Since 1948, TAGA, Quebec City, Canada, pp. 528–542.
R. Asikainen, (2010). Quality analysis of a printed
natural reference image, Master’s thesis, Aalto
University, Finland
C. Temponi, F. D. Fard, H. W. Corley, (1999) A fuzzy
decision model for color reproduction, International
Journal of Production Economics 58 (1), 31–37.
A. Verikas, M. Bacauskiene, (2005) Image analysis and
fuzzy integration applied to print quality assessment,
Cybernetics and Systems 36 (6) 549–564.
J. Tchan, R. C. Thompson, A. Manning, (1999) A
computational model of print-quality perception, Expert
Systems with Applications 17, 243–256.
L. Breiman, (2001) Random forests, Machine Learning
45, 5–32.
L. Breiman, J. H. Friedman, R. A. Olshen, C. J. Stone,
(1993) Classification and Regression Trees, Chapman &
Hall,.
ISSN: 2231-5381
http://www.ijettjournal.org
Page 326