Suggested Research Topics
The following research topics suggested by Drs. Tappert and Cha are suitable for D.P.S. or M.S. dissertations,
and also for briefer preliminary studies in research-related courses.
Of course, the research conducted for a D.P.S. dissertation must be more extensive than that for an M.S. dissertation,
and similarly for an M.S. dissertation relative to a small research study required for a course.
For many of these topics preliminary research has been undertaken either by us or by some of our students.
Significant research opportunities exist in each of these topics
by either extending the earlier work or by undertaking new approaches.
There is a clear distinction between research and project work.
Research is original, rigorous work that advances knowledge, improves professional practice,
and/or contributes to the understanding of subject.
Research methods depend upon the nature of the research: controlled experiment, empirical studies,
theoretical analyses, or other methods as appropriate.
We require DPS research work to be of sufficient strength to be able to distill from it a paper worthy
of publication in a refereed journal or conference proceedings.
Project work, on the other hand, uses known technology to develop systems,
usually according to specified customer requirements.
Many of our projects, however, are developed to provide support for our research,
so there is interplay between the project and research activities.
For an overview of research and projects in these areas,
see Research/Projects Interplay
and associated slides.
We have four major areas of research –
biometrics, interactive visual image studies, natural language processing and related forensics studies,
and general pattern recognition studies –
and several other areas that, depending on the particular study, tend to relate to one or several of these major areas.
The four areas are also highly interrelated –
biometrics, for example, is basically a subarea of pattern recognition.
For listings of various studies in these areas,
see Dr. Tappert's publication list
and Dr. Cha's publication list
For an overview of this area, see
An introduction to biometric recognition by Jain, et al., 2004
or the textbook used in our DPS course on Emerging Information Technologies.
The main problem we consider is the task of establishing the distinctiveness of each individual in a population
when there is a set of measurements that have an inherent variability for each individual.
This task of establishing individuality can be thought of as showing the distinctiveness
of the individual classes with a small error rate in discrimination.
This is important for most forensic science applications such as writer, face, iris, fingerprint, speaker,
or bite mark identification.
All these applications face the problem of scientifically establishing individuality,
which is motivated by court rulings such as Daubert vs. Merrell Dow Pharmaceuticals
that concerns the uniqueness of handwriting.
Dr. Cha's dissertation, for example, was on writer individuality where he employed a dichotomy model
of feature vector differences that is inferable to the general population.
There are other biometric applications for which studies can be made using the same model,
and preliminary studies have been made on some of these (see below).
Individuality based on keystroke patterns
We have explored the keystroke biometric for long text input which had previously received little attention.
Three students have completed DPS dissertations in this area.
Masters-level projects have extended the keystroke biometric studies in two ways –
authentication studies using Cha's dichotomy method and longitudinal studies –
The following are possible additional studies using the keystroke biometric data:
- Develop a new general procedure or methodology
and apply it to the keystroke data as a test case.
- Explore user authentication (binary decision - yes, you are the person you claim to be, or no, you are not)
using Cha's dichotomy method, perhaps comparing against other methods, and using the keystroke data as a vehicle.
The DPS dissertations so far have explored the identification problem (1-of-n decision),
so exploring authentication is a natural extension.
- Further explore longitudinal studies in general or perhaps
Cha's dichotomy method in the context of longitudinal studies.
- A study of the degree to which increasing user populations lower accuracy.
This could be done for recognition or authentication and could use the keystroke data as a vehicle.
We have given this area some attention, for example see Biometric system evaluation studies below.
- Other classification techniques could be applied to the data since
the currently used nearest neighbor technique is one of the simplest available.
- Other optimization techniques could be investigated for various aspects of the keystroke system:
fallback procedure, feature extraction, outlier removal, etc.
- Studies could be designed to determine the biometric value of having users correct or edit their keyboard input.
We have conducted preliminary investigations of speaker individuality based on earlier work
by Dr. Cha on writer individuality. The idea is to determine whether a speaker's voice is
sufficient to accurately verify the identity of the speaker.
Related work involves, for example, such topics as whether a speaker can disguise his/her speech
sufficiently so as not to be identified.
See the CSIS Research Day 2004 paper summarizing an M.S. Dissertation, Jan 2004,
Establishing the Uniqueness of the Human Voice for Security Applications.
Individuality based on iris images
See the CSIS Research Day 2004 paper summarizing a study by M.S. Research Seminar student Seung Choi,
Use of Histogram Distances in Iris Authentication
and the 2005 Int. J. Graphics, Vision and Image Processing article,
On the Individuality of the Iris Biometric
Individuality based on mouse movement patterns
Mouse movement patterns also have the potential for identification or verification.
One could study, for example, mouse movements and mouse clicks for performing the standard editing operations
of insert, delete, copy, etc.
The mouse is used for moving the cursor to highlight text in cutting,
to mark target locations in pasting, and to click on dropdown menus.
Editing methods can further differentiate users -- for example, some use dropdown menus, others use shortcut keys.
Several masters-level projects have initiated work in this area, see CSIS Research Day papers.
Individuality based on writing style (stylometry)
The words, syntax, etc. used in writing can be used to identify or verify the author.
Several masters-level projects have initiated work in this area, see CSIS Research Day papers.
Although not biometric, for a related study that counts word usage, see
He Counts Your Words.
Individuality based on fingerprints
On-line fingerprint verification
by Jain, et al., used Dr. Cha's dichotomy technique to investigate the individuality of fingerprints.
Biometric system evaluation studies
We are beginning to research various methods of evaluating biometric systems.
Our first study in this area is a paper (work performed by an M.S. student) entitled
Evaluation of Biometric Identification in Open Systems
presented at the Audio- and Video-based Biometric Person Authentication (AVBPA 2005) Conference.
Another general problem is determining how well a system developed on a small population of users
generalizes to larger populations.
For example, Dr. Cha proposed in his dissertation a dichotomy model of feature vector differences
that is potentially inferable to the general population.
It would be interesting to conduct further studies, of either a theoretical or empirical nature, on this or other models.
Interactive Visual Image Research
Combining human and machine capabilities can lead to powerful systems.
In this regard we are inspired by a quote by Albert Einstein who said:
"Computers are incredibly fast, accurate and stupid; humans are incredibly slow, inaccurate and brilliant;
together they are powerful beyond imagination."
We are interested in improving accuracy and speed in visual recognition tasks.
Specifically, we want to enhance human-computer interaction in applications of pattern recognition
where higher accuracy is required than is currently achievable by automated systems,
but where there is enough time for a limited amount of human interaction.
This topic has so far received only limited attention from the research community.
Our current, model-based approach to interactive recognition was originated at RPI,
and then investigated jointly at RPI and Pace University (see Interactive flower recognition below).
Our NSF Proposal
gives a good introduction and background to this area.
The proposal's objective was to develop guidelines for the design of mobile interactive object classification systems;
to explore where interaction is, and where it is not, appropriate;
and to demonstrate working interactive recognition systems.
The proposal targeted three very different domains – foreign signs, faces, and skin diseases –
and this approach could also be used in other domains.
Interactive flower recognition
Our initial success in recognizing flowers established a methodology for continued work
in the area of interactive visual systems, see
Interactive flag recognition
An M.S. student, Eduardo Hart, completed studies resulting in two conference papers:
Interactive Flag Identification Using Image Retrieval Techniques
Interactive Flag Identification Using a Fuzzy Neural Technique.
Interactive skin lesion recognition
See the study by M.S. student John Sikorski:
Identification of Malignant Melanoma by Wavelet Analysis.
Interactive rare coin grading
Completed 2003 DPS dissertation by Rick Bassett entitled
Computer-based Objective Interactive Numismatic System.
Interactive archeological artifact studies
Completed 2005 DPS dissertation by Sheb Bishop entitled
Classification of Greek Pottery Shapes and Schools Using Image Retrieval Techniques.
Interactive analysis of paintings
Completed 2005 DPS dissertation by Tom Lombardi entitled
The Classification of Style in Fine-Art Painting, and related conference papers.
Natural Language Processing and Related Forensics Research
Pen computing studies involving shorthand and chatroom symbols
Alphabets of shorthand symbols have been developed for handheld devices, notably the Graffiti alphabet
for Palm handhelds. An extension of this for even faster input would be to use shorthand symbols
that correspond to words and phrases such as the text symbols used for chatroom communication.
See the summaries of an M.S. in CS dissertation: the IWFHR 2004 conference paper
Use of Chatroom Abbreviations and Shorthand Symbols in Pen Computing
and the HCI 2005 conference paper
Common Chatroom Abbreviations Speed Pen Computing.
This work could be extended.
Automatic extraction of dynamic information from static images of handwriting for forensic studies
Forensic handwriting examination has the challenge of working only with static images of handwriting.
However, much of the information relating to the identity of the specific writer,
or relating to whether the writing was natural or distorted, is contained in the dynamics of the handwriting.
Therefore, it can be important in the field of forensic science to develop digital image processing techniques
to evaluate the dynamic and temporal components of the handwriting from static images.
Handwriting style analysis
Handwriting originates from a particular copybook style such as Palmer or Zaner-Bloser that one learns in childhood.
Since questioned document examination plays an important investigative and forensic role in many types of crime,
it is important to develop a system that helps objectively identify a questioned document's handwriting style.
Identifying the copybook style of a questioned document, such as a ransom note, can help to reduce the scope
of the suspect population in the identification of an individual writer.
Preliminary offline (OCR type) work on this subject resulted in the conference paper,
Similarity-Based Handwriting Copybook Style Identification
and the Journal of Forensic Document Examiners article
Handwriting Copybook Style Identification for Questioned Document Examination.
In 2005 Mary Manfredi extended this work using pseudo online data to complete her DPS dissertation entitled
"Copybook Style Determination of Pseudo-Online Handwriting Data,"
summarized in a paper presented at the International Graphonomics Society (IGS 2005) conference,
Similarity-Based Copybook Style Analysis Using Pseudo-Online Handwriting.
Handwriting synthesis of a particular writer's style
This topic is related to the handwriting style analysis described above.
Forensic examiners would like the capability of creating a handwriting document in a particular writer's style.
For example, the FBI might want to synthetically create a ransom note in a particular writer's style.
The detection of handwriting forged by novices
It is known that many forgeries, particularly those by novice forgers, are
written slowly in order to accurately capture the writing shape and style of the true writer.
Initial work in this area dealt with the development of a fractal number estimate of the wrinkliness
of the handwriting from static images
where we found an inverse correlation between the wrinkliness and the speed of the writing.
See the summary of an M.S. dissertation presented at the International Graphonomics Society (IGS 2004) conference,
The Detection of Forged Handwriting Using a Fractal Number Estimate of Wrinkliness.
This research could be extended is several ways.
An analyzer that determines a writer's handedness and pen grip from static images of the handwriting
For forensic examinations it is important to be able to determine
the probability that a writer of a document was left-handed or right-handed.
Determining the writer's pen grip is also of significant value.
Language recognition from short telephone speech input
Completed 2002 DPS dissertation by Jonathan Law entitled
An Efficient First Pass of a Two-Stage Approach for Automatic Language Identification of Telephone Speech.
Language recognition from text input
This is a topic that has important applications but has been minimally explored.
One application is to find websites in a particular language and possibly in a particular domain as well.
Another application is to detect a language shift, say from English to French,
within the same document
with the purpose of appropriately shifting to the proper accent when converting
into speech in a Text-To-Speech (TTS) system.
Otherwise, the TTS system when reading such sentences will treat everything as though it were English
and likely mispronounce words in the other language.
Working in this area, Bashir Ahmed, DPS 2004, completed his dissertation entitled
"Detection of Foreign Words and Names in Written Text," summarized in the CSIS Research Day 2004 paper
Language Identification from Text Using N-gram Based Cumulative Frequency Addition.
See CSIS Research Day paper describing preliminary work by an M.S. student from a course assignment,
A Neural Network Classifier for Junk E-Mail.
This work is currently being extended by DPS 2006 student Ted Markowitz,
see CEAS 2006 Conference Paper.
Association Rules in Microarray Data
See paper describing preliminary work by an M.S. student,
Mining Association Rules in Microarray Data.
General Pattern Recognition Studies
There are many problems in the area of pattern recognition,
and several applications of pattern recognition have been described in the biometrics and interactive visual image sections.
Applications to other areas are possible, and
more fundamental pattern recognition problems can also be investigated.
Fundamental Pattern Recognition Research
Fundamental pattern recognition research can involve such areas as feature extraction,
similarity and distance metrics, and pattern classification techniques.
For example, see the following:
A theoretical study of shape recognition
Completed 2006 DPS dissertation study by Carl Abrams,
see conference papers.
AI and Pattern Recognition studies relating to the Human Brain
See CSIS Research Day papers describing masters-level projects in this area.
Also see Introduction to Neural Networks.
Other Suggested Research
Technology life cycle and business-related studies
In the second-year DPS course on emerging information technologies,
we studied the phases of the technology life cycle as described by Kendall and Kurzweil.
Such a dissertation, for example, could examine (compare/contrast) the phases of the cycles
described by Kendall and Kurzweil (and possibly other versions of the technology life cycle),
give examples of technologies as they have gone through the cycle,
and present one or more hypotheses and provide evidence to support them.
Your hypothesis might be that the time period from invention to general use has been getting shorter and shorter,
as might be anticipated by Kurzweil's idea of time speeding up.
Another hypothesis might be that the time period from invention to general use for Internet inventions
is shorter than for non-Internet inventions.
Combinatorial optimization problem
This is the problem of assigning students (or company employees) to project teams
where each student (or employee) lists his/her project preferences,
preferences for team meeting locations (based on where they live),
preferences for scheduling weekly team meetings (day/time), and work experience.
The instructor (or company project manager) can give different weights to the four parameters (and possibly others)
and specify a different number of students to each project.
The algorithm should optimize the assignment of students to team projects based on the above (or other) constraints,
and each team should have a good mix of team member work experiences (leadership as well as technical capability).
For preliminary work on this problem see the
team group assignment paper.
This is an NP-hard problem that is amenable to sub-optimal optimization algorithms, such as genetic algorithms.
If you are interested in this problem, you can work with us, or possibly with Lixin Tao or Mike Gargano.
"Look Who's Speaking" Robot
For interactive man/machine dialogue it is desirable to develop a robot
that turns to focus on the speaker.
This will involve the determination of the location of the speaker
from the audio signal of the voice source or from the visual image of the speaker, or both.