What is Rex?
Interface explanation
How it works
Code
Credits
Acknowledgments
Onomastic


What is Rex?

Rex is a browser for written documents

You select a writing feature and browse documents by similarity. The presently available dataset is IAM OffLine 3.0.

Rex is a showcase for a document retrieval method

Rex is distinct from the common content-based image retrieval systems in two significant regards:

1. you interact verbally with the machine, i.e. the document is described by selecting a feature and a numerical value, instead of presenting the system with a reference image;

2. interaction is based on perceptual features, not mathematical concepts, thus permitting browsing by lay persons.

Rex is a graphonomics teaching tool and knowledge resource on writing characteristics of specific populations

For each document a profile in form of a plot is provided that allows comparison between documents and sampled populations. See also this visualization of the entire IAM OffLine 3.0 dataset sampled in the canton of Bern, Switzerland, on about 600 writers and 1539 documents.

Interface explanation

Feature – A perceptual characteristic of script, such as slant, some defined here.

Instrument – The computational instrument used to characterize the feature. Rex uses orientation (the orientation of the writing contour at each pixel location) and run length (the horizontal "white" distance between two "ink" strokes in each row of image pixels).

Measure – The mathematical concept underlying the perceptual feature and the numerical value allowing the ranking of documents.

Writing image – A colorcoded representation of the contour orientation of the writing (red is for horizontal strokes). Roll over the image to get an enlargement, and click on the enlargement to see the original document.

Writing profile – The orientations or run lengths represented as a plot.

How it works

Read about the contour orientation feature and its applications to graphonomics in these papers:

V. Atanasiu, L. Likforman-Sulem, and N. Vincent, "Writer Retrieval – Exploration of a Novel Biometric Scenario Using Perceptual Features Derived from Script Orientation," in Proc. 11th Intl. Conf. on Document Analysis and Recognition, Beijing, China, 2011, pp. 628-632.

V. Atanasiu, "Forensic vs. Computing writing features as seen by Rex, the intuitive document retriever," Proc. First Intl. Workshop on Automated Forensic Handwriting Analysis, Beijing, China, 2011, pp. 16-20.

V. Atanasiu, L. Likforman-Sulem, N. Vincent, "Talking Script. Retrieval of written documents by description of script features," Gazette du Livre Medieval, to be published.

Code

Code for obtaining contour orientation profiles is made available as Matlab source code and Windows executable (you need to install the Matlab Runtime Component before running Rex proper).

Versions log:
2011.08.11 – First public release

Credits

Rex was coded by Vlad Atanasiu during his stay in Laurence Likforman's Handwriting Computing Laboratory, at Telecom ParisTech, Paris, France, in 2010-2011.

Contacts: Vlad Atanasiu, atanasiu@alum.mit.edu, http://alum.mit.edu/www/atanasiu/

Acknowledgments

We like to thank the dataset providers for allowing the use of their work in conjunction with Rex. We also want to extend our gratitude to Laurence Likforman for having enable this work come into being.

Onomastic

Rex is the king of document retriever dogs to which you can talk in human language about the documents you want him to fetch you.