Resif : On-line Handwriting Recognition by Hierarchical Fuzzy Inference Systems

Contact : Eric Anquetil ( eric.anquetil@irisa.fr )

RESIF technology is currently composed of three main softwares to analyze, model and recognize handwritten characters and words:

  • RESIFCar is the software specialized to recognize handwritten characters: latin alphabet, digits and special symbols.

  • RESIFMot is the software for unconstrained cursive handwritten words recognition.

  • RESIFApp is the automatic learning process that generates from a handwritten character database the hierarchical fuzzy models used by the recognition systems: RESIFCar and RESIFMot.

RESIFCar and RESIFApp are yet in their fourth version. RESIFCar is a writer independent on-line handwriting recognition system able to deal with cursive handwritten characters. It is the result of our last 10 years of research. In 2001, through an industrial collaboration with PurpleLabs company http://www.purplelabs.com ResifCar has been successfully integrated into mobile devices (smartphones) which can be characterized by their limited computing and memory resources.

 

The smartphone was based on ARM 7 TDMI microprocessor. In the context of this practical implementation, the recognizer must compute into a device with limited resources : 13 Mhz CPU, RAM size of 50 Kbytes and ROM size (program and data) around 200 Kbytes available for this feature. Moreover, the input method which combines LCD display with a digitizer is based on a low tactile screen definition (0.20 mm pixel pitch) with a sampling frequency of 20 points per second. These mobile phones are today commercialized in Europe.

 

 

Work is in progress to extend the capability of the recognition system RESIFMot which is today in a beta version. The aim is to optimize lexical post-processing to deal with large vocabulary and to reduce computation time and memory resources.

Recent videos are available here.