an important link between two industries—papermaking
and printing—and would contribute to further quality
improvements of both paper and prints.
This paper presents an example of such a system devel-
oped to assess, explore, and monitor print quality in offset
colour printing using data collected from the two indus-
tries. The core of the system is a set of virtual sensors oper-
ating on images acquired on-line from the printing press.
The outputs from these sensors are soft measures–inferen-
tial calculations—of print quality. The developed software
of the system allows predicting values of these measures,
called quality attributes (QAs), using parameters charac-
terising the paper manufacturing and printing processes
as input variables to the prediction model. Random forests
[17] is a core technique used for modelling. In addition to
predicted values, random forests also provide an estimate
of the variable importance—a measure of impact the vari-
ables have on model accuracy.
1.1. Virtual sensors in printing industry
There are numerous examples of virtual sensors applied
in both papermaking and printing industries [18]. Trepa-
nier et al. have demonstrated that image processing can
be used to estimate paper surface characteristics such as
unevenness of a paper sheet [19]. The sensor has the
advantage of capturing spatial information in two dimen-
sions which would be lost when using conventional meth-
ods. Surface unevenness affects ink transfer distribution
and has impact on print quality [20]. Ink density has also
been measured with sufficient accuracy using image pro-
cessing and soft computing techniques [21]. A system able
to detect regions suitable for measuring colour registry, i.e.
misalignment of printing plates has been proposed [22].
Template matching systems able to detect regions and col-
ours exposed for distortion in the printed page were devel-
oped [7,23]. A quality control system developed by
Shankar et al. for flexo-gravure printing also is based on
the template matching approach [24].
A good example of a virtual sensor for estimating a print
quality attribute is the study of Sadovnikov et al. where a
method to estimate mottling in colour prints was proposed
[25]. Virtual sensors aimed at assessing quality of halftone
rasters have been studied in [26]. The authors showed how
irregularity of a halftone dot pattern could be measured
using three different types of soft measures based on coef-
ficients of the 2D Fourier transform.
1.2. Soft computing in printing industry
Three generations of systems targeted for improving
print quality or making the printing process more efficient
can be distinguished. The categorisation is based on the
ability of the system to monitor, explore and explain vari-
ous quality attributes.
The first generation systems are able only to monitor
some print quality attributes from images acquired on-
line. These systems lack the ability of utilising the non-for-
malised knowledge that operators possess [22,24,27].
The second generation systems have the ability to ex-
ploit the operator knowledge to some extent. Perner pre-
sents an example of such a system [23], where image
processing techniques and operator prior knowledge are
used to handle a detected print quality defect. Evans and
Fisher used decision trees to extract print operator knowl-
edge in the form of explicit rules [28]. Surprisingly enough
even knowledge not obvious for the operators was de-
tected by the expert system. Perhaps the most obvious
example of learning from press operators is the work by
Almutawa and Moon, where the system learned to mimic
operator actions when adjusting ink flow [2]. A neural net-
work based controller was developed by Englund for con-
trolling ink flow based on online analysis of double grey-
bar images, see Fig. 2 [29].
The third generation systems are capable of utilising
information not only from the press and operators but also
information concerning properties of paper. It is believed
that exploitation of such information may lead to higher
quality of prints. To our knowledge, there have been only
two attempts to use some paper parameters in print qual-
ity modelling [30,31]. Utilization of soft computing tech-
niques such as artificial neural networks, fuzzy systems,
case-based reasoning, and decision trees, is a distinctive
feature of the second and third generation systems. It is
worth mentioning that most of the earlier studies have fo-
cused on flexographic and ink jet printing.
2. Approach
The proposed system is capable of assessing print qual-
ity attributes from images acquired on-line in a printing
press and connecting these quality attributes with param-
eters of both paper manufacturing and printing processes.
In contrast to previous studies relying on detection of suit-
able spots, where an image is acquired and evaluated, the
proposed system uses designated halftone areas known
as double grey-bars, shown in Fig. 2.
Such grey-bars are used in ordinary production for the
grey balance control—for balancing the CMY inks to repro-
duce a neutral grey. This is done by manual inspection by
operators eyes or using densitometers often following a
calibration methodology such as G7 [32]. Grey-bars are
common in world lithographic newsprint and are typically
printed at the edge of each page, and are, therefore, desir-
able to be used as measurement spots for automatic print
Fig. 2. Double grey-bar used to measure quality attributes of offset print.
J. Lundström et al. / Measurement 46 (2013) 1427–1441
1429