Demo on Optical Character Recognition (OCR) for handwritten numerals.
models (dir) - Different SVM models trained for OCR system.
collect_chars - Collecting training examples for OCR.
mpaper - Allows to enter handwritten characters by mouse.
ocr_fun - Classifies numerals and displays result.
save_chars - Saves images to file.
train_ocr - Trains OCR classifier based on multiclass SVM.
tune_ocr - Tunes SVM for OCR problem.
ocrmodel.mat - SVM model used by OCR system.
Description:
First try
demo_ocr
The demo_ocr allows to draw hand-written numerals from 0 to 9 and
classify them by multi-class SVM. The standard computer mouse
is used as the input device.
The complete set of function necessary to design the OCR are
included. To train your own OCR proceed as follows:
- Collecting training set. Use 'collect_chars' to draw your
examples of numerals or other characters. Then save all training
examples to a directory DataDir.
- Tuning SVM parameters. Customize the file 'tune_ocr' or
use already tuned SVM paramater.
- Training SVM model: Customize the file 'train_ocr' and run it.
- Run OCR demo 'demo_ocr'.
About: Statistical Pattern Recognition Toolbox
(C) 1999-2003, Written by Vojtech Franc and Vaclav Hlavac
Czech Technical University Prague
Faculty of Electrical Engineering
Center for Machine Perception
Modifications:
05-jun-2004, VF
04-jun-2004, VF
10-sep-2003, VF