Logo image
Performance Evaluation of 2D Shape Features for Pattern Classification
Thesis

Performance Evaluation of 2D Shape Features for Pattern Classification

Chia-Han Chien
Masters, 國立清華大學, 資訊工程學系
2004

Abstract

二維形狀 分類 輪廓 區域 動量 2D Shape classification contour region moment
Object shape features are powerful when used in similarity search and retrieval. In a recognition system, feature extraction and classifiers play an important role. We just focus on both feature extraction and classifiers. In this thesis, we discuss two major ways to extract shape features: region-based shape, contour-based shape. The methods of extracting shape features include moment invariants, Zernike moments, Pseudo Zernike moments, Angular Radial Transform (ART), improved moments, and Fourier descriptors. Then we introduce a classifier SVM. A comparison with different feature extraction for different classifiers will be tested on 3 databases composed of 252 128x128 images, such as DB1, DB2, DB3 from PRIP Lab at NTHU. Finally, we show the simulation results and make a conclusion.

Metrics

1 Record Views

Details

Logo image