Feature Selection Using Neural Network In Writer Identification Domain

Goh , Hui Li (2011) Feature Selection Using Neural Network In Writer Identification Domain. Project Report. UTeM, Melaka, Malaysia. (Submitted)

	PDF (24 Pages) Feature_Selection_Using_Neural_Network_Network_In_Writer_Identification_Domain-24_Pages.pdf - Submitted Version Download (3MB)
	PDF (Full Text) Feature_Selection_Using_Neural_Network_Network_In_Writer_Identification_Domain-Full_Text.pdf - Submitted Version Restricted to Registered users only Download (11MB)

Abstract

Handwriting is different for different individuals. However, every single person has their own writing styles which are the unique features that are underlying in their writing. Yet, not every feature is important in identifying the writer. In order to identify these relevant features, feature selection technique is used. Today, feature selection has been used in handwriting aspect using different approaches but not neural networks. In this study, neural network will be used for feature selection in identifying w1iters in order to classify the handwritten into an appropriate class label by using Neural Network as classifier. Throughout the study, a three layered MLP architecture will be used as the neural net structure and Visual C++ will be used to develop the program for the research. The research is basically referred to the technical paper which entitled ·'Neural Network Feature Selector", but it has been modified to adapt to this problem situation. As a conclusion, the proposed research in this study has fulfilled the objectives. Feature selection using neural network has performed well in writer identification domain in this study. However, there is still room for improvement to make it better and computation friendly.

Actions (login required)

View Item