| Summary: | 碩士 === 國立高雄海洋科技大學 === 電訊工程研究所 === 99 === Biometrics is a common topic. In this field, neural network is a common machine learning algorithm, and it has been applied to many fields. Recently, support vector machines (referred to as SVM) which is based on statistical learning theory catches the most attention. It is because SVM has the better recognition capability and faster calculation speed than the general neural networks; furthermore, it does not have the situation of over-learning. There are many researches proving that SVM has good performance of recognition in the open literature.
Probabilistic neural network (referred to as PNN) is a kind of neural network based on Bayesian decision theory, and it belongs to the feedforward network architecture. PNN is highly regarded due to its short training time, and also, it does not have the iterative process. In this thesis, we apply in human face recognition and Traditional Chinese handwriting recognition.
Most researches use the public face databases in human face recognition. For example, they are ORL, Yale, INDIAN, etc. Thus, we use data source both from ORL and the database created by ourselves in this study. In handwriting, the database was made of 20 persons handwriting in Traditional Chinese. In this study, we consider individual handwriting habits use different quantitative methods to explore the feasibility of using handwriting recognition as an identification identity.
The experimental results show that the recognition rate by using SVM and ORL is 96%, while the recognition rate for our database is 92%(Block Background) and 80%(White Background) respectively. In Traditional Chinese handwriting, the best rate of using SVM to recognize is 75%, while the best rate for PNN is 80%.
|
|---|
