Principal Component Analysis Dimensionality Reduction For Writer Verification

Ramlee, Rimashadira (2015) Principal Component Analysis Dimensionality Reduction For Writer Verification. Masters thesis, Universiti Teknikal Malaysia Melaka.

[img] Text (24 Pages)
Principal Component Analysis Dimensionality Reduction For Writer Verification.pdf - Submitted Version

Download (506kB)

Abstract

Writer verification (WV) is a process to verify whether two sample handwritten document are written by the same writer or not. WV also known as one to one comparison process, where the process is more specific which compare one writer to another writer. Therefore, this process needs a unique characteristic of the writer in order to prove the owner of the handwritten document. Basically, different person will have different type of handwriting styles usually it is unique between each other. Furthermore, most of the previous research in handwriting analysis field was used the unique characteristic to represent the individuality of handwriting. A part from that, individuality of handwriting became main issue in this study in order to fulfill requirement of WV process. In previous verification framework of WV the individuality of handwriting was acquired by using feature extraction process. Meanwhile, previous verification framework of WV consists of Preprocessing task, feature extraction task and classification task. In this study, using the previous verification framework are not enough to produce the best result in verification process. This is because the quality of individuality of handwriting that has been acquired is less effective in representing the uniqueness of the writer. Therefore, this study was proposed Dimension reduction technique for acquiring the individual features of the handwritten data henceforth improved the previous verification’s framework in order to enhance the verification accuracy. The sample data was taken from IAM online database which this database is the benchmark for handwriting analysis research. Five writers with 3619 instance of images are chosen for the experiment whereas 9 documents of handwriting samples are taken from each writer and more than 50 word randomly divided into training and testing dataset. Both dataset is will be process by Principal Component Analysis which is one of the dimension reduction techniques. PCA was applied after feature extraction process whereas the reduction process will resulted low dimensional of new subspace of data. By using the data resulted by PCA the classification process by random forest was conducted in order to verify the writer of the handwritten document. The individuality representation is implemented by presenting various representations of individual feature into more important feature are selected by using the proposed technique to be used in verifying the writer. Experimental show that the performance of the proposed methods has improved the verification rate of 90.00 % and above overall of the result with the reduction is successful in each data set. However, overall of the result the improved framework still cannot verify 100 % accurately the writer of the handwritten data.

Item Type: Thesis (Masters)
Uncontrolled Keywords: Computer vision, Writing, Identification, Writer Verification
Subjects: H Social Sciences > H Social Sciences (General)
H Social Sciences > HV Social pathology. Social and public welfare
Divisions: Library > Tesis > FTMK
Depositing User: Mohd Hannif Jamaludin
Date Deposited: 04 Aug 2016 04:25
Last Modified: 11 Nov 2020 08:37
URI: http://eprints.utem.edu.my/id/eprint/16858
Statistic Details: View Download Statistic

Actions (login required)

View Item View Item