Continuous Moment-Based Features for Classification of Ground Vehicle SAR Images

Abstract

In this paper, four continuous moment-based feature extraction techniques for Synthetic Aperture Radar (SAR) images are examined. Geometric Moments (GMs), Legendre Moments (LMs), Zernike Moments (ZMs) and Pseudo Zernike Moments (PZMs) are introduced as a feature extraction for three types of ground vehicles from SAR images. GMs are simplest moment that suffers from high degree of information redundancy since its basis is not orthogonal. LMs defined as a moment with orthogonal basis to overcome GMs drawback. Complex moments are defined as ZMs and PZMs and widely used because their polynomials are orthogonal to each other and are rotational invariant. However, PZMs have better feature representation capabilities than ZMs based method. In this context, we applied the four techniques on SAR images using Support Vector Machine (SVM) for classification. Experimental results have proven that the accuracy of ZMs and PZMs are superior to GMs and LMs, while LMs still has a better accuracy rather than GMs.