Krawtchouk Moment Invariant and Gaussian ARTMAP Neural Network: a combination techniques for image classification

Abstract

The main objective of this research is to develop a practical system for binary image classification using Krawtchouk Moment Invariant (KMI) as the feature extraction technique while Gaussian ARTMAP (GAM) is adopted for classification task. Fundamentally, KMI is introduced by P.T. Yap back in 2003 based on the discrete orthogonal function which is invariant to position, scale and rotation factors. This technique is used to extract the global shape feature of binary images. As a comparison we also applied two other types of features extraction methods that are Geometric Moment Invariant (GMI) and Legendre Moment Invariant (LMI). In doing so, 20 dissimilar types of insect with totally of 240 images have been used for classification purposes. Furthermore, we have applied k-folds cross validation technique in order to seek the reliability of the techniques used. In this research, we found that KMI generated the highest classification rate of GAM which is about 99% compare to GMI (91%) and LMI (97%). The high share numbers for KMI, GMI and LMI demonstrated that GAM neural networks is well efficient technique for classification. In addition, the combination of GAM and KMI methods is one of the brilliant concepts in developing a fully practical system for binary image classification based on the global shape features.