Handwritten Character Recognition

Exhibitor: Miguel Po-Hsien Wu

Supervisor: Zhao Dong

Research Group: Complex and Intelligent Systems

Industry Sector:

Character recognition plays an important role in the modern world. It can solve more complex problems and makes humans’ job easier. An example is handwritten character recognition. This is a system widely used in the United States to recognize zipcode or postal code for mail sorting.

There are different techniques that can be used to recognize handwritten characters. Two techniques researched in this thesis are: Pattern Recognition and Artificial Neural Network (ANN). Both techniques are defined and different methods for each technique is also discussed. Bayesian Decision theory, Nearest Neighbor rule, and Linear Classification or Discrimination are types of methods for Pattern Recognition. Shape recognition, Chinese Character and Handwritten Digit recognition uses Neural Network to recognize them.

Some of the advantages of using Neural Networks for recognition are: more like a real nervous system, can solve problems with multiple constraints, it is insensible to noise, often good for solving complex problems.

Neural Network is used to train and identify written digits. After training and testing, the accuracy rate is 99%. This accuracy rate is very high.

The system has more trouble identifying numeral “5”. This is probably caused by the fact that the numeral is not fully connected or running together. The image-file created from the net-file does not show a clear numeral that was trained. This should be improved in the future.

Thesis Document (PDF)