PREFERENCES OF HANDWRITING RECOGNITION ON
MOBILE INFORMATION SYSTEMS IN MEDICINE
Improving Handwriting Algorithm on the Basis of Real-life Usability Research
Andreas Holzinger, Martin Schlögl
Institute of Medical Informatics, Statistics and Documentation, Research Unit HCI4MED, Medical University Graz
Graz, Austria
Bernhard Peischl
Institute of Software Technology, Graz University of Technology, Graz, Austria
Matjaz Debevc
Faculty of Electrical Engineering and Computer Science, University of Maribor, Maribor, Austria
Keywords: Handwriting Recognition, Mobile Computer, Human-computer Interaction, Usability, Real-life,
Health Care.
Abstract: Streamlining data acquisition in mobile health care in order to increase accuracy and efficiency can only
benefit the patient. The company FERK-Systems has been providing health care information systems for
various German medical services for many years. The design and development of a compatible front-end
system for handwriting recognition, particularly for use in ambulances was clearly needed. While
handwriting recognition has been a classical topic of computer science for many years, many problems still
need to be solved. In this paper, we report on the study and resulting improvements achieved by the
adaptation of an existing handwriting algorithm, based on experiences made during medical rescue
missions. By improving accuracy and error correction the performance of an available handwriting
recognition algorithm was increased. However, the end user studies showed that the virtual keyboard is still
the overall preferred method compared to handwriting, especially among participants with a computer usage
of more than 30 hours a week. This is possibly due to the wide availability of the QUERTY/QUERTZ
keyboard.
1 INTRODUCTION AND
MOTIVATION FOR RESEARCH
In cases of emergency, rapid patient information
collection is very important. This information is
most often collected by first aiders (first responders)
and paramedics (e.g. Red Cross). Prompt and
accurately recorded and well communicated vital
patient data can make the difference between life
and death (Holzman, 1999), (Anantharaman & Han,
2001).
The data acquisition should have as little
disruptive effect on the workflow of the emergency
responders (rescue staff) as possible. A possible
solution for data input can be an mobile application
on a lightweight handheld device (Baumgart, 2005),
(Chittaro, Zuliani & Carchetti, 2007).
Due to the fact that emergencies are usually
within difficult physical situations, special attention
to the design of information technology for
emergencies has to be taken into consideration
(Klann et al., 2008). A key issue of any such
information system is the acquisition of textual
information. However, extensive text entry on
mobile devices is principally to be avoided and a
simple and easy to use interface, in accordance with
the proverb: less is more, is a supreme necessity
(Holzinger & Errath, 2007).
14
Holzinger A., Schlögl M., Peischl B. and Debevc M. (2010).
PREFERENCES OF HANDWRITING RECOGNITION ON MOBILE INFORMATION SYSTEMS IN MEDICINE - Improving Handwriting Algorithm on the
Basis of Real-life Usability Research.
In Proceedings of the International Conference on e-Business, pages 14-21
DOI: 10.5220/0002979900140021
Copyright
c
SciTePress
The basic evidence that entering data onto a
mobile device via a stylus is slower, more erroneous
and less satisfactory for end users than entering data
via a QWERTZ (de) or QUERTY (us) keyboard has
been demonstrated in some studies (Haller et al.,
2009), although, on the other hand the use of a stylus
is much faster and more accurate than using finger
touch (Holzinger et al., 2008b). A specific study for
“Ambulance Run Reporting” shows good results for
acquiring text with a virtual keyboard, while
acquiring text by the application of handwriting
recognition showed some serious usability problems
(Chittaro et al., 2007). Motivated by this previous
work, we focus in this work on handwriting
recognition and on how to improve its usability – in
case of need, also by adaptation of existing
handwriting algorithms. Consequently, in this paper
we report on real-life experiences and on some
improvements achieved by the adaptation of an
existing handwriting engine.
2 BACKGROUND
A big difficulty of handwriting recognition is that
handwritten characters are variable on an individual
basis and that these characters are usually separated
into alphabets, numerals, and symbols, despite the
different characters of the language itself. Although
handwriting recognition will benefit in future from
improved adaptive and context-sensitive algorithms,
improving the user experience of novice end users
with the respective technology is possibly the most
important factor in enhancing user acceptance
(MacKenzie & Chang, 1999). This is even more
important in medical or health care contexts, where
the difficulty is in the environmental conditions, e.g.
if the person is on the move or in a hurry (Holzinger
et al., 2008a). Whereas the first problem might be
solved by the training modus opportunities, in order
to adapt the system to the individual handwriting
style, the second problem is only solvable by an
extremely robust and usable system. Especially in
the health care domain, good end user acceptance
and usability can only be obtained by providing
simple operation (good user guidance), very short
response times and low error rates (Holzinger,
Geierhofer & Searle, 2006).
Basically, there are several methods for
handwriting recognition; these belong basically to
two distinct families of classification:
I) Structured and Rule Based Methods
Because of the fuzzy nature of human handwriting,
it makes sense to adapt the well known fuzzy logic
technique for this purpose (Gader et al., 1997).
Rather than evaluating the two values as in digital
logic, fuzzy terms admit to degrees of membership
in multiple sets so that fuzzy rules may have a
continuous, rather than stepwise, range of truth of
possibility. Therefore non-identical handwritten
numerals, from same or different users, can be
approximated using fuzzy logic for fast and robust
handwriting recognition (Shi & Li, 2006).
II) Statistical Methods
a) Hidden Markov Modeling (HMM)
The attractiveness of HMM for various pattern
recognition tasks is mainly due to their clear and
reliable statistical framework. Many efficient
algorithms for parameter estimation and model
evaluation exist, which is an important prerequisite
for their practical implementation for real-life
applications (Plotz & Fink, 2009). The methods
using HMM (Marti & Bunke, 2002), are based on
the arcs of skeleton graphs of the words to be
recognized and an algorithm applied to the skeleton
graph of a word extracts the edges in a particular
order, which is transformed into a 10-dimensional
feature vector. Each of these features represent
information about the location of an edge relative to
four reference lines, the curvature and the degree of
the nodes incident to the considered edge. Training
of the HMM is done by use of the Baum-Welch
algorithm, while the Viterbi algorithm is used for
recognition (Bunke, Roth & Schukattalamazzini,
1995), (Xue & Govindaraju, 2006).
b) Neural Networks
The methods based on Neural Networks were driven
by the emergence of portable, pen based computers.
A typical approach is to combine an artificial neural
network (ANN), as a character classifier, with a
context-driven search over segmentation and word
recognition hypotheses (Yaeger, Webb & Lyon,
1998).
However, handwriting recognition not only consists
of the recognition itself; the data must undergo some
preprocessing:
(I) Reduce noise;
(II) Normalization, and
(III) Segmentation.
The last step, the segmentation phase, segments the
input into single characters (Plamondon & Srihari,
2000). Writing discrete characters requires no
segmentation; this is done by the users themselves
(Tappert, Suen & Wakahara, 1990).
PREFERENCES OF HANDWRITING RECOGNITION ON MOBILE INFORMATION SYSTEMS IN MEDICINE -
Improving Handwriting Algorithm on the Basis of Real-life Usability Research
15
Another way to improve recognition is to
decrease the set of possible alternatives, such as to
restrict the set to accepting only lower case letters or
digits (Frankish, Hull & Morgan, 1995).
3 RELATED WORK
To date only a few studies considered handwriting
recognition on mobile devices and very few in the
health care domain.
A very early work by Citrin et al. report very
general on the usage of a pen on a flat surface of a
LCD unit (scribing and tapping). They reported that
with the maximum rate of 100 selections of direction
per second for pen, scribing may produce strokes
with the speed of 300 (100×3) bps. However, no
more results were found (Citrin et al., 1993).
MacKenzie showed that the recognition accuracy
for a set containing upper and lower case letters was
lower than for a set containing just lower case letters
(MacKenzie et al., 1994).
Chittaro evaluated a system for recording data on
a system during a running ambulance drive, having
first responders as participants. Text entry via virtual
keyboard and handwriting recognition (MS
Transcriber – Calligrapher) were also performed.
Text entering by handwriting was considered very
laborious and difficult by the users (Mean 3.8, Var
6.6), while entering text by use of the virtual
keyboard was quite easy (Mean 7.2, Var 1.8).
(0=Hard, 9=Easy). Furthermore, they emphasized
the bad usability of entering text by using
handwriting recognition. Most words were wrongly
recognized and there were enormous problems in
correcting those wrongly recognized words (Chittaro
et al., 2007).
4 METHODS AND MATERIALS
The aim of our study was to increase the
performance of available handwriting recognition by
improving accuracy and error correction following
solid usability engineering methods (Holzinger,
2005).
We focused on separate character recognition,
since the correction of a single letter, at the moment
of false recognition, can be made more naturally,
and efficiently, than attempting to correct or delete a
single letter within a recognised word.
Due to limited space, there could be some
problems inputting long words. Therefore, only one
character at a time can be written and recognized.
4.1 Experimental Device
The device used for the prototype was an Asus
MyPad A626 PDA (Personal Digital Assistant).
This device is equipped with an anti-glare touch
screen display. For typing on the touch screen, a
stylus is used.
Table 1 contains the technical specifications of this
device.
Table 1: Specifications of the PDA ASUS
MyPal A626.
CPU
Marvell XScale, 312MHz
Operating System
MS Windows® Mobile™ 6
Memory
256MB Flash ROM and 64 MB
SDRAM
Display
3.5" Brilliant TFT LCD
65k full-colours, anti-glare
16-bit display QVGA,
240x320 px
touch screen
Weight
158g
Physical
dimensions
117 mm x 70.8 mm x 15.7cm
4.2 Dialog Design
Figure 2: Design of the handwriting dialog.
The light green area within the writing sections
defines the optimal size for handwritten lowercase
characters of 80 points (Phatware, 2002).
4.3 Handwriting Recognition
We used the SDK of the handwriting recognition
engine Calligrapher (in MS Windows® Mobile
Transcriber) in the version 6.0 (Phatware, 2002).
ICE-B 2010 - International Conference on e-Business
16
This SDK makes it possible to define single
character recognition. We can handle the results and
a custom timeout (after which time the recognition
starts) can be defined.
4.3.1 Adaptive Timeout
A handwritten character consists of one or more
strokes. The recognition starts after the character is
finished. The system has to await a timeout before
starting recognition because the system doesn’t
know whether the character consists of just one or
more strokes.
A stroke is defined as the writing from pen down
to pen up (Tappert et al., 1990).
Because of the different writing speeds of each
user, this timeout has to be calculated for each user.
Therefore, the system stores the last ten times which
elapse between two strokes.
Figure 3: Calculation timeout T [sec].
Figure 2
shows how the timeout is calculated every
time a timeout is requested. s(1) is the last calculated
average time between strokes, s(2)..s(11) are the last
ten stored times between strokes. X is a factor, in
this experimental setting X is 200. The result T is the
timeout in seconds.
4.3.2 Correction Intervention
Calligrapher SDK 6.0 doesn’t adapt recognition on
users’ handwriting because of the use of static
Fuzzy-Neuronal Nets (Strenge, 2005).
There are problems with some user’s style of
writing letters – the user writes a letter (e.g. an “a”)
but the recognition engine recognizes another letter
(e.g. figure 4).
Figure 4: Written “a” but not recognized as “a”, instead as
“ir”.
A recognition result is a list of possible characters
and its weight (maximum 5 entries). Every time the
same letter is wrongly recognized for a user (as in
Figure 2), the lists returned by the recognition are
similar.
These lists (characters and its weight) with its
representing letter are stored. Each of them is called
schema.
During writing, the recognition result will be
compared to the stored schemas as follows.
(Example in Figure 5)
For each stored schema:
Characters from the result list and the list of the
schema are compared. If the result list consists of 2
or 3 characters, at least 2 have to match to the stored
schemas lists characters. (2 of 2, 2 of 3). If there are
4 or 5 characters in the result list, at least 3 have to
match (3 of 4, 3 of 5). This means, the resulting list
is validated to the list of the schema.
If the list is valid according to the list of the
schema, the average deviation between these
matching characters is calculated.
Figure 5: Example of a list comparison.
The representing letter of the schema with the lowest
average deviation will be put in first place of the
recognition result.
4.3.3 Calibration
The calibration is designed to collect user specific
data for each letter. This data contains weights,
which present every character explicitly. Also,
schemas of wrongly recognized letters (Chapter
3.3.2) are collected.
The system prompts the user to input a letter.
If the result list of the recognition has the prompted
letter in first place, the weight will be stored for this
letter. In the calibration phase, at least 2 weights will
be stored for each letter.
If not, the result list will be stored as a schema with
the prompted letter as a representing letter. In the
calibration phase, a maximum of 10 schemas for
each letter is stored.
This calibration is done once for each user. A
continuous calibration is also done during writing in
the handwriting recognition dialog, saving weights
and schemas for correctly recognized letters (but not
for deleted letters)
PREFERENCES OF HANDWRITING RECOGNITION ON MOBILE INFORMATION SYSTEMS IN MEDICINE -
Improving Handwriting Algorithm on the Basis of Real-life Usability Research
17
4.3.4 Other Interventions on Recognition
Results
To avoid side effects, the intervention described in
Chapter 3.3.2 is only made when the weight of a
recognized letter is less than the average weight for
this letter (average of the weights for this letter
collected by calibration).
Other interventions are made to avoid potential.
problems with highly confusable pairs such as “r”
and “v” (Frankish et al., 1995). (I) While writing a
word, only letters and punctuation marks are valid,
recognized results. (II) Just deleted letters (with
BACKSPACE) are not valid, recognized results for
the next recognition (III) Special handling for “O”
and “0” as first letter of a word or number
4.4 Experiment
The real life environment is mostly a seat in an
ambulance car. To avoid negative effects on
ambulance responder’s work, the experiment is done
in their recess in the ambulance service rooms,
simulating the circumstances (sitting in a car) by
doing the experiment sitting on a chair, holding the
PDA in their hand, without laying down the elbows
on e.g. an armrest. (Kjeldskov et al., 2004) shows
that simulating environments gives almost the same
results.
Figure 6: Participants during experiments in real life.
Participants were people who work as ambulance
officers (professionals, volunteers and former
civilian service). No previous experience with
mobile computers was required.
They were asked to fill out a background
questionnaire to obtain data about their age,
education and use of computers.
The prototype for the experiment is divided into two
parts, one for virtual keyboard based text input, and
the other for handwriting recognition input. Within
these two parts, the users have the opportunity to
become familiar with the input methods. After that,
the user has to input a given text to the experimental
dialog (for measuring the accuracy). Due to
measuring the accuracy, text entry is done as text
copy (MacKenzie & Soukoreff, 2002). This text
consists of 13 German words (94 characters without
spaces, 106 with spaces). After the keyboard based
experimental dialog, the calibration of the
handwriting is done.
Speed in wpm, words per minute (Lewis, 1999,
MacKenzie et al., 1994) and the accuracy of the
handwriting recognition are measured and
calculated. At the end, a feedback questionnaire is
filled out by the user. Some questions are based on
the study of Chittaro (Chittaro et al., 2007).
5 RESULTS
5.1 Participants
The participants of the experiment were professional
(9) and volunteer (8) first responders of the Austrian
Red Cross, one student of medicine and three others
(because everyone could be a volunteer first
responder).
10 are experienced on a PDA or a mobile phone
with touch screen, while 11 have no experience with
touch screens.
Their ages ranged from 20 to 85 years. Two
elderly people (68 and 85 years) were chosen
because they had never before used a QWERT
keyboard or a PC.
The average use of a PC is 12.3 years, using a
PC 31 hours per week. 11 participants use a PC 30
hours a week, while 10 participants use a PC for
more than 30 hours.
One of the 21 participants was left-handed.
5.2 Accuracy
Overall 30 weekly
usage
> 30 weekly
usage
Mean Var Mean Var Mean Var
99.1 6.28 100 11.5 99.06 1.44
Figure 6: Accuracy inputting text with virtual keyboard
[%]; all participants, participants 30 hours and above.
Overall
30 weekly
usage
> 30 weekly
usage
Mean Var Mean Var Mean Var
89.25 34.3 91.43 30.20 88.00 37.34
ICE-B 2010 - International Conference on e-Business
18
Figure 7: Recognition accuracy [%] of handwriting
recognition; all participants, participants 30 hours and
above with interventions.
Overall
30 weekly
usage
> 30 weekly
usage
Mean Var Mean Var Mean Var
84.66 57.6 86.99 79.15 83.33 38.21
Figure 8: Recognition accuracy [%] of handwriting
recognition; all participants, participants 30 hours and
above without interventions.
The participants using a PC 30 hours a week
include the two elderly people.
The 85 year old participant has an accuracy of
89.2% for inputting text with the virtual keyboard
and a recognition accuracy of 80.1% with
interventions and 65.6% without interventions.
The 68 year old participant had an accuracy of
100% for inputting text with the virtual keyboard
and a recognition accuracy of 95% with
interventions and 90.8% without interventions.
The 85 years old participant has an accuracy of
89.2% for inputting text with the virtual keyboard
and a recognition accuracy of 80.1% with
interventions and 65.6% without interventions.
The 68 year old participant has an accuracy of
100% for inputting text with the virtual keyboard
and a recognition accuracy of 95% with
interventions and 90.8% without interventions.
5.3 Speed
Overall
30 weekly
usage
> 30 weekly
usage
Mean Var Mean Var Mean Var
13.17 27.7 12.88 29.46 13.43 18.29
Figure 9: Words per minute virtual keyboard; all
participants, participants 30 hours and above.
Overall
30 weekly
usage
> 30 weekly
usage
Mean Var Mean Var Mean Var
8.44 4.59 8.11 5.37 8.71 1.95
Figure 10: Words per minute handwriting recognition; all
participants, participants 30 hours and above.
Participants using a PC 30 hours a week include
two elderly people.
The 85 year old participant wrote 2.87 wpm with
the keyboard and 2.82 wpm with handwriting
recognition. The 68 year old participant wrote 4.88
wpm with the keyboard and 4.17 wpm with
handwriting recognition.
5.4 User Questionnaire
Overall
Mean Var
Keyboard
Inputting Data (+4=easy, -4=difficult) 3.0 2.6
Correction of wrong inputted data
(+4=easy, -4=difficult)
4.0 2.8
Handwriting
Inputting Data (+4=easy, -4=difficult) 2.0 4.9
Correction of wrongly
input/recognized data (+4=easy, -
4=difficult)
3.0 1.4
Did the recognition slow down your
writing (+4=no, -4=yes)
0.5 9.2
I would prefer (+4=handwriting, -
4=keyboard)
-2.5 7.9
Basic Information
Use of colour is (+4=useful, -
4=useless)
2.0 2.5
The handwriting recognition positively
surprised me (+4=yes, -4=no)
2.5 7.1
Characters on the PDA are easy to read
(+4=yes, 4=no)
4.0 3.9
Figure 11: Overall results of user questionnaire.
Weekly
computer usage
[hours]
Mean
(<=30)
Var
(<=30)
Mean
(>30)
Var
(>30)
Keyboard
Inputting Data 3.5 1.4 3.0 3.7
Correction of
wrongly input
data
4.0 1.7 3.5 4.5
Handwriting
Inputting Data 2.5 4.2 0.5 3.8
Correction of
wrongly
input/recognized
data
4.0 0.5 2.5 1.8
Did the
recognition slow
down your
writing
2.5 8.4 -0.5 8.2
I would prefer -1.5 2.0 -2.5 4.9
Basic Information
Use of colour 2.0 2.5 1.5 2.6
The handwriting
recognition
positively
surprised me
4.0 7.7 0.0 4.7
PREFERENCES OF HANDWRITING RECOGNITION ON MOBILE INFORMATION SYSTEMS IN MEDICINE -
Improving Handwriting Algorithm on the Basis of Real-life Usability Research
19
Characters on the
PDA are easy to
read
4.0 4.6 4.0 3.2
Figure 12: Results of user questionnaire for weekly usage
of computer 30 hours and above.
6 CONCLUSIONS
Entering text with the virtual keyboard (Mean 3.0,
Var 2.6) was easier for the participants than with
handwriting (Mean 2.0, Var 4.9). However,
compared to the study of (Chittaro et al., 2007), we
could reach an significant improvement by inputting
data with handwriting. Interestingly, inputting data
by handwriting recognition was rated easier by
participants who use computers less than or equal to
30 hours a week than by participants with
extensively more use (Mean 2.5; Var 4.2; against
Mean 0.5, Var 3.8 of virtual keyboard). Also, the
correction on the handwriting recognition dialog was
rated easier (Mean 4.0, Var 0.5; against Mean 2.5,
Var 1.8; of virtual keyboard). Participants with a
computer usage of more than 30 hours a week
preferred the virtual keyboard (Mean -2.5, Var 4.9)
more than the other participants (Mean -1.5, Var
2.0). This could be a result of hardly any
handwriting during work and much more typing text
on classical keyboards (QWERTZ or QUERTY).
Consequently, the two elderly participants were
included in this study, in order to obtain data
regarding participants who never used any computer
or handheld device. The elderly participants were
the only ones who provided a complete preference to
the handwriting recognition in contrast to the virtual
keyboard. This is also clearly visible in the results
for these participants, although both groups have
quite comparable results in wpm for the virtual
keyboard and the handwriting text input.
This is an interesting result; however, it is not of
practical relevance, since there are hardly any people
left – at least amongst people able to volunteer as a
first responder – without experience on computer
keyboards. Today, from elementary school on,
children get used to work with computers by using
the QWERTZ or QUERTY keyboard.
Nevertheless, our interventions on the basis of
the results of the handwriting recognition, finally
paid off in an significant improvement on the
recognition accuracy (over all participants a better
accuracy of Mean +4.39%, Var 9.54).
These interventions can also be useful for the
improvement of other handwriting recognition
engines, due to the fact that our interventions were
only made on the results of the engine, achieving
better accuracy. The use of a handwriting
recognition engine with a higher accuracy than e.g.
Calligrapher, in combination with our demonstrated
interventions, may even improve the overall
accuracy. Our methods on operating on the results of
the handwriting recognition engine operate context
independent. Using a dictionary to add the
likelihood of upcoming characters may improve the
accuracy in that part of the problem regarding
confusable pairs, such as “r” and “v”. Because of
typing in characters one by one, a word completion
feature could be added to handwriting recognition
too. This also would increase the writing speed.
ACKNOWLEDGEMENTS
This study was performed with support of FERK-
Systems. We cordially thank the engineering team
for their continued and effective industrial support of
this work. The research was partially funded by the
Austrian Research Promotion Agency (FFG) within
one „Innovationsscheck Österreich“.
REFERENCES
Anantharaman, V. & Han, L. S. (2001) Hospital and
emergency ambulance link: using IT to enhance
emergency pre-hospital care. International Journal of
Medical Informatics, 61, 2-3, 147-161.
Baumgart, D. C. (2005) Personal digital assistants in
health care: experienced clinicians in the palm of your
hand? The Lancet, 366, 9492, 1210-1222.
Bunke, H., Roth, M. & Schukattalamazzini, E. G. (1995)
Off-Line Cursive Handwriting Recognition Using
Hidden Markov-Models. Pattern Recognition, 28, 9,
1399-1413.
Chittaro, L., Zuliani, F. & Carchetti, E. (2007) Mobile
Devices in Emergency Medical Services: User
Evaluation of a PDA-based Interface for Ambulance
Run Reporting. In: Löffler, J. & Klann, M. (Eds.)
Mobile Response. Berlin, Heidelberg, New York,
Springer, 19-28.
Citrin, W., Halbert, D., Hewitt, C., Meyrowitz, N. &
Shneiderman, B. (1993) Potentials and limitations of
pen-based computers. Proceedings of the 1993 ACM
conference on Computer science. Indianapolis,
Indiana, United States, ACM.
Frankish, C., Hull, R. & Morgan, P. (1995) Recognition
accuracy and user acceptance of pen interfaces.
Conference on Human Factors in Computing Systems.
Denver, Colorado, United States, ACM
Press/Addison-Wesley Publishing Co.
ICE-B 2010 - International Conference on e-Business
20
Gader, P. D., Keller, J. M., Krishnapuram, R., Chiang, J.
H. & Mohamed, M. A. (1997) Neural and fuzzy
methods in handwriting recognition. Computer, 30, 2,
79-86.
Haller, G., Haller, D. M., Courvoisier, D. S. & Lovis, C.
(2009) Handheld vs. Laptop Computers for Electronic
Data Collection in Clinical Research: A Crossover
Randomized Trial. Journal of the American Medical
Informatics Association, 16, 5, 651-659.
Holzinger, A. (2005) Usability Engineering for Software
Developers. Communications of the ACM, 48, 1, 71-
74.
Holzinger, A. & Errath, M. (2007) Mobile computer Web-
application design in medicine: some research based
guidelines. Universal Access in the Information
Society International Journal, 6, 1, 31-41.
Holzinger, A., Geierhofer, R. & Searle, G. (2006)
Biometrical Signatures in Practice: A challenge for
improving Human-Computer Interaction in Clinical
Workflows. In: Heinecke, A. M. & Paul, H. (Eds.)
Mensch & Computer: Mensch und Computer im
Strukturwandel. München, Oldenbourg, 339-347.
Holzinger, A., Hoeller, M., Bloice, M. & Urlesberger, B.
(2008a). Typical Problems with developing mobile
applications for health care: Some lessons learned
from developing user-centered mobile applications in
a hospital environment. International Conference on
E-Business (ICE-B 2008), Porto (PT), IEEE, 235-240.
Holzinger, A., Höller, M., Schedlbauer, M. & Urlesberger,
B. (2008b). An Investigation of Finger versus Stylus
Input in Medical Scenarios. ITI 2008: 30th
International Conference on Information Technology
Interfaces, June, 23-26, 2008, Cavtat, Dubrovnik,
IEEE, 433-438.
Holzman, T. G. (1999) Computer-human interface
solutions for emergency medical care. interactions, 6,
3, 13-24.
Kjeldskov, J., Skov, M. B., Als, B. S. & Høegh, R. T.
(2004) Is It Worth the Hassle? Exploring the Added
Value of Evaluating the Usability of Context-Aware
Mobile Systems in Field. Mobile HumanComputer
Interactoin - MobileHCI2005. Springer Berlin /
Heidelberg, 529-535.
Klann, M., Malizia, A., Chittaro, L., Cuevas, I. A. &
Levialdi, S. (2008) HCI for emergencies. CHI '08
extended abstracts on Human factors in computing
systems. Florence, Italy, ACM.
Lewis, J. R. (1999). Input Rates and User Preference for
three small-screen input methods: Standard Keyboard,
Predictive Keyboard and Handwriting. Human
Factors and Ergonomics Society.
MacKenzie, I. S. & Chang, L. (1999) A performance
comparison of two handwriting recognizers.
Interacting with Computers, 11, 3, 283-297.
MacKenzie, I. S., Nonneke, B., Riddersma, S., McQueen,
C. & Meltz, M. (1994) Alphanumeric entry on pen-
based computers. International Journal of Human-
Computer Studies, 41, 5.
MacKenzie, I. S. & Soukoreff, R. W. (2002) Text Entry
for Mobile Computing: Models and Methods, Theory
and Practice. HUMAN-COMPUTER-INTERACTION,
17, 2, 147–198.
Marti, U. V. & Bunke, H. (2002) Using a statistical
language model to improve the performance of an
HMM-based cursive handwriting recognition systems.
Hidden Markov models: applications in computer
vision. World Scientific Publishing Co., Inc., 65-90.
Phatware (2002) Calligrapher SDK 6.0 Developer's
Manual.
Plamondon, R. & Srihari, S. N. (2000) On-Line and Off-
Line Handwriting Recognition: A Comprehensive
Survey. IEEE Transactions on Pattern Analysis and
Machine Intelligence, 22, 1, 63-84.
Plotz, T. & Fink, G. A. (2009) Markov models for offline
handwriting recognition: a survey. International
Journal on Document Analysis and Recognition, 12, 4,
269-298.
Shi, B. & Li, G. (2006) VLSI Neural Fuzzy Classifier for
Handwriting recognition. Patent US 7,146,037.
Strenge, M. (2005) Konzepte und Toolkits zur
Handschrifterkennung.
Tappert, C. C., Suen, C. Y. & Wakahara, T. (1990) The
State of the Art in On-Line Handwriting Recognition.
IEEE Transactions on Pattern Analysis and Machine
Intelligence, 12, 8, 787-808.
Xue, H. H. & Govindaraju, V. (2006) Hidden Markov
models combining discrete symbols and continuous
attributes in handwriting recognition. Ieee
Transactions on Pattern Analysis and Machine
Intelligence, 28, 3, 458-462.
Yaeger, L. S., Webb, B. J. & Lyon, R. F. (1998)
Combining neural networks and context-driven search
for on-line, printed handwriting recognition in the
Newton. Neural Networks: Tricks of the Trade. Berlin,
Springer-Verlag Berlin, 275-298.
PREFERENCES OF HANDWRITING RECOGNITION ON MOBILE INFORMATION SYSTEMS IN MEDICINE -
Improving Handwriting Algorithm on the Basis of Real-life Usability Research
21