Susan E. George

On-line handwritten signature verification using wavelets and back-propagation neural networks

This paper investigates dynamic handwritten signature verification (HSV) using the wavelet transform with verification by the backpropagation neural network (NN). It is yet another avenue in the approach to HSV that is found to produce excellent results when compared with other methods of dynamic, or on-line, HSV. Using a database of dynamic signatures collected from 41 Chinese writers and 7 from Latin script we extract features (including pen pressure, x and y velocity, angle of pen movement and angular velocity) from the signature and apply the Daubechies-6 wavelet transform using coefficients as input to a NN which learns to verify signatures with a False Rejection Rate (FRR) of 0.0% and False Acceptance Rate (FAR) less of than 0.1.

Online Pen-Based Recognition of Music Notation with Artificial Neural Networks

Computer Music Journal, 27:2, pp. 70–79, Summer 2003 2003 Massachusetts Institute of Technology. There are many different reasons why we might want to enter music notation into a computer, including editing and composing tasks, educational music theory exercises, sophisticated searching of music archives based on melodic fragments supplied as search criteria, or simply producing transposed arrangements. Musicologists may be interested in analyzing the style of a collection, and copyright enforcers may be interested in detecting legal infringements. At present, musical information is most often entered into the computer using the computer keyboard, a mouse, a piano keyboard (or some other electronic instrument) attached to the computer, or a scanning device for a sheet of printed music with offline recognition of printed symbols. Various optical musical recognition (OMR) techniques have been developed to convert scanned pages of music into a machine-readable format. Blostein and Baird (1992) presented a critical survey of problems and approaches to music image analysis. Since then, work in the OMR field has continued with researchers such as Bainbridge and Carter (1997) and Bainbridge and Wijaya (1999), who wrote a system to convert optically scanned pages of music into a machine-readable format. More recently, Droettboom and Fujinaga (2001) created an adaptive music-recognition system and interpretation mechanism. Fujinaga and Riley (2002) concentrated upon recommendations and options of file formats in the context of creating an archival image containing all relevant data extracted from a printed score. This makes interpretation of the music for archival storing, web delivery, printing, and other applications possible. Other researchers have focused upon novel process techniques, such as Ng (2002), who recognized that a stroke-based segmentation approach using mathematical morphology is necessary in OMR, applied after the image pre-processing (i.e., thresholding, de-skewing, and basic layout analysis). Similar image-processing techniques have been explored for handwritten music (Roach and Tatem 1988; Ng 2001). In particular, the work conducted by Luth (2002) focused on the recognition of handwritten music manuscripts. It is based on imageprocessing algorithms like edge detection, skeletonization, and run-length. In addition, general image-processing methods have been explored that are applicable to both printed and handwritten music (George forthcoming). With online music input, the prevailing interface technology for music editing involves the conventional ‘‘point-and-click’’ mouse-based paradigm. The ‘‘point-and-click’’ method typically requires various musical symbols to be selected from a menu and meticulously placed on a staff, and a constant movement between the menu and the staff is necessary. A parallel in the context of word processing would be to individually select the alphabetical characters and place them on the page to compose a sentence. In other fields that require some written notation (whether signatures or postcodes, mathematical notations, or cursive writing), online input has moved away from ‘‘point-and-click’’ approaches to penbased recognition. However, with few exceptions, there is a noticeable absence of research into the pen-based recognition of music symbols. One common paradigm is simply using the pen as a stylus to select music symbols from a menu bar. In another paradigm, the user must learn a special sequence of movements to enter a given music symbol. The most desirable system, however, would allow the user to write conventional music symbols.

Statistical modelling of general practice medicine for computer assisted data entry in electronic medical record systems

Electronic medical record (EMR) systems have much potential, however, there are still a number of issues that need to be resolved before EMRs are widely accepted. One of these issues is the data input task, a potentially serious practical barrier to on-line medical computer usage. This paper reports the empirical modelling of data input requirements for physicians who use a problem-orientated medical record system. Three statistical models (Bayesian conditional probability, multiple linear regression and discriminant analysis) to predict drug treatment given problem diagnoses are derived from EMRs of 2500 general Practice encounters. Two metrics are used to measure the predictive power of the models considering both the number of drugs correctly predicted and the strength with which the models predict them. The models are tested on 500 unseen records from the same patient-physician population and the data used to build the models. The Bayesian model produces the best predictions on unseen data and is also the easiest model to compute. A prototype interface that enables new patient cases to be entered is constructed to demonstrate how the predictive power of the model can translate into benefits in the data entry task.

Believe It or Not: Virtual Religion in the 21st Century

This paper considers the development of virtual communities focusing upon virtual religion and its impact on humanity. It is important that religion is expressed communally and socially, and the Internet has provided a new community context for religiosity, linking people from geographically, socially, and culturally disparate backgrounds, facilitating interactivity as never before in an intriguing anthropological development. We find examples of “online religion†that are already occurring and see technology as playing a crucial positive role in humanity in the 21st century. While there are limitations with virtual interaction compared to face-to-face engagement, there are benefits, primarily that such technology starts to answer the deeper philosophical questions associated with technology, especially the question of how technology cannot rob people of the essence of what it is to be human. By facilitating virtual religion, technology assists the uniquely human pursuit of religiosity, and merely provides a new “meeting place†for exchange.

Spatio-temporal analysis with the self-organizing feature map

Abstract. Spatio-temporal pattern recognition problems are particularly challenging. They typically involve detecting change that occurs over time in two-dimensional patterns. Analytic techniques devised for temporal data must take into account the spatial relationships among data points. An artificial neural network known as the self-organizing feature map (SOM) has been used to analyze spatial data. This paper further investigates the use of the SOM with spatio-temporal pattern recognition. The principles of the two-dimensional SOM are developed into a novel three-dimensional network and experiments demonstrate that (i) the three-dimensional network makes a better topological ordering and (ii) there is a difference in terms of the spatio-temporal analysis that can be made with the three-dimensional network.

Biometric verification in dynamic writing

Pen-tablet devices capable of capturing the dynamics of writing record temporal and pressure information as well as the spatial pattern. This paper explores biometric verification based upon the dynamics of writing where writers are distinguished not on the basis of what they write (ie the signature), but how they write. We have collected samples of dynamic writing from 38 Chinese writers. Each writer was asked to provide 10 copies of a paragraph of text and the same number of signature samples. From the data we have extracted stroke-based primitives from the sentence data utilizing pen-up/down information and heuristic rules about the shape of the character. The x, y and pressure values of each primitive were interpolated into an even temporal range based upon a 20 msec sampling rate. We applied the Daubechies 1 wavelet transform to the x signal, y signal and pressure signal using the coefficients as inputs to a multi-layer perceptron trained with back-propagation on the sentence data. We found a sensitivity of 0.977 and specificity of 0.990 recognizing writers based on test primitives extracted from sentence data and measures of 0.916 and 0.961 respectively, from test primitives extracted from signature data.

Neural unit shape representation: A new SOM-based visualisation

The Self Organising Map (SOM) is an artificial neural network technique that has been successfully applied in clustering and visualisation tasks in data mining. In this paper, we propose a new SOM-based visualisation, Neural Unit Shape Representation. Its advantage over the previous SOM-based visualisation, U-matrix and Component Planes, is that it gives visual cues to the properties of individual neural units or nodes on the map, enabling direct examination of the produced ordering. Further, the combination of the Neural Unit Shape Representation and a new way of drawing boundaries using U-matrix gives a good visualisation and helps people, even without prior knowledge of data, to find interesting patterns.