The aim of the PAPYRUS project was developing pen-based application environments and end-user applications which feature a natural person/computer interface via the usage of the electronic paper and the recognition of unrestricted handwriting and gestures.
SCRIPTOR is a recognition engine of dynamic cursive script, acquired through an electronic paper device, that can be trained to a specific writing style, in order to facilitate incremental learning. The input is a file of raw data and the output is a file of word alternatives, in order of decreasing confidence. SCRIPTOR consists of 5 processes that communicate via message passing in a UNIX environment.
Firstly, the preprocessing, normalization, and codification module. The raw data are divided into strokes according to the speed profile of the pen. Each stroke is represented by means of a code of 9 integers.
Secondly, the preliminary recognition module, that transforms the sequence of coded strokes into a table of partial matches to be used to produce the output recognition file.
Thirdly, a module that integrates the graphic information stored in the table of partial matches with linguistic knowledge stored in a dictionary. It operates by successive approximations, looking at the number of strokes and at islands of confidence in the table in order to restrict the search in the dictionary.
Fourthly, a module that allows the allographic databases to be updated during the normal use of the system in order to achieve the functionality of incremental learning. The user operates as a critic in a reinforcement learning paradigm. No intervention is interpreted as a positive feedback as recognition operation. An explicit intervention is required to correct the error of the system.
Fifthly, a module that manages the database of allographs, considered as self organized neural networks of prototypes. Hebbian learning is used for tuning the network and an original technique has been implemented for controlling its dynamic growth of the network.
Many people still rely on pen and paper to carry out their daily work either because they find a computer too difficult to use of insufficiently mobile.
Research efforts are devoted to investigating:
methods, algorithms and techniques for online recognition of cursive scripts and for processing language related data;
architectures suitable for the implementation of these techniques;
electronic paper devices and gestural man machine interfaces.
In carrying out its work, PAPYRUS draws on the results of other ESPRIT projects and academic studies in handwriting recognition.
The project contributed to the advance in the basic technologies needed to pursue this aim. Research effort was thus spent in the fields of: algorithms, methods and techniques for the recognition of dynamic cursive script and gestures; neural computing; methods and techniques for language analysis; integration of cursive script recognition in pen-based software platforms; pen-based work group systems; pen-based applications.
CB2 5LQ Cambridge
6500 HE Nijmegen
NG1 4BU Nottingham
NG3 1GE Nottingham