Improving Triangle Geometry Shape Features Through Triangle Points Selection In Digit Recognition

Arbain, Nur Atikah (2016) Improving Triangle Geometry Shape Features Through Triangle Points Selection In Digit Recognition. Masters thesis, Universiti Teknikal Malaysia Melaka.

	Text (24 Pages) Improving Triangle Geometry Shape Features Through Triangle Points Selection In Digit Recognition.pdf - Submitted Version Download (87kB)
	Text (Full Text) Improving Triangle Geometry Shape Features Through Triangle Points Selection In Digit Recognition.pdf - Submitted Version Restricted to Registered users only Download (4MB)

Abstract

Geometry features has been widely used in image processing especially in face recognition, fingerprint recognition, digit recognition, vehicle detection and also in intrusion. Among the commonly used geometry features are the features that are based on triangle properties. Generally, triangle properties can be used to produce the features for image classification. To produce these features, triangle geometry need to be formed based on three coordinates which are the corners of A, B and C. However, not all triangle formations can be formed from the three coordinates due to the condition where corners of A, B and C may cause a straight line problem. The straight line problem occurs when the chosen coordinates of the corners of A, B and C are in a straight line which causes the triangle geometry impossible to be formed. On the other hand, the straight line occurs when the gradient of corners A, B and C produces the equivalent value. This can be proved by the experiment conducted to identify the gradient that has equivalent value where the position of coordinates A, B and C will determine either the triangle can be formed or vice versa. The purpose of this study is to suggest an improvement on triangle geometry shape through triangle point selection. To achieve this purpose, there are two objectives suggested for this study. They are: i) to propose straight line detection technique for corner A, B and C of triangle; and ii) to improve triangle shape by proposing location of corners based on dominant distribution of foreground image. In the experiment, four types of digit dataset are chosen which are IFCHDB, HODA, MNIST and BANGLA where each datasets is consisted of testing data and training data. The Detection of Triangle Point Selection (DTPS) is proposed to detect the triangle point that caused a straight line to be formed. Then, the straight line problem is solved using Triangle Geometry Based Dominant Distribution of Foreground Image (TD2FI). Next, the Triangle Features Based Summation of Gradient and Ratio (TSGR) and Enhancement of Proposed Triangle Features using Absolute Value (EFTA) are proposed in order to improve the classification accuracy result. The experimental results are yielded by comparing the results of classification accuracy between the present proposed methods with a prior proposed method using the supervised machine learning (SML). The SML used are the Support Vector Machine (SVM) and the Multi-Layer Perceptron (MLP). The result of classification accuracy has shown impressive results for TD2FI, TSGR and EFTA methods through the SVM and MLP techniques whereas the datasets from IFCHDB, HODA and BANGLA respectively have acquired good results through the SVM technique while MNIST dataset has acquired the highest result of classification accuracy through the MLP technique. The result of classification accuracy for TD2FI is 94.723% from IFCHDB dataset, 97.295% from HODA dataset, 95.4% from MNIST dataset and 90.275% from BANGLA dataset. In conclusion, the proposed method is capable of outstripping the straight line issue based on the position of the coordinates of corners A, B and C as well as produce a better result for classification accuracy.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Image processing, Digital techniques, Pattern recognition systems
Subjects:	T Technology > T Technology (General) T Technology > TA Engineering (General). Civil engineering (General)
Divisions:	Library > Tesis > FTMK
Depositing User:	Muhammad Afiz Ahmad
Date Deposited:	31 Mar 2017 01:30
Last Modified:	10 Oct 2021 15:35
URI:	http://eprints.utem.edu.my/id/eprint/18357
Statistic Details:	View Download Statistic

Actions (login required)

View Item