OCR Algorithms Jacek Bajor
OCR Optical Character Recognition Translation of printed or handwritten text to digital form KAZ 813
OCR Fields Pattern Recognition Artificial Intelligence Machine Learning Computer Vision
Applications Digitalising libraries
Applications books.google.com
Applications Office work
Applications Captchas Mr. blocked
Applications Recognising licence plates And many, many others...
OCR Algorithms Artificial Neural Networks
Why ANN?
Why ANN? They are flexible  They can be thought They can learn themselves They are powerful
ANNs Let’s assume ANN is a black magical box
ANNs Learning stage input output a a a a a a
ANNs Evaluation 98% a 5% a?
ANN structure Neuron (nerve cell)
ANN structure
ANN structure
ANNs types Feedforward Recurrent (back-propagation) Radial basis function network Kohonen self-organizing network Others...
ANNs applications Optical Character Recognition Function  approximation Artificial Intelligence Finance Speech analysis Many, many others
Advantages Flexible Efficient if well designed
Disadvantages Focused on one activity Give approximations Inefficient if poorly designed
Libraries FANN - Fast Artificial Neural Network Library - leenissen.dk/fann Annie - annie.sourceforge.net Libann - savannah.nongnu.org/projects/libann
Thank you

OCR