Image Analysis Using Disc-Harmonic Moments and Their RST Invariants in Pattern Recognition

Moments and moment invariants are the most useful tools in pattern recognition. Recently, the Conventional Disc-Harmonic Moments (CDHMs) are used to describe binary and gray scale images. In order to deal with color images in a holistic manner, these CDHMs are generalized as Quaternion Disc-Harmonic Moments (QDHMs) by using the quaternion algebra. Then the Rotation, Scaling and Translation (RST) invariants (CDHMIs and QDHMIs) are derived for more description of images that have undergone affine transformations. In this paper we first illustrate the discrimination power of these moments by evaluating their efficiency in image reconstruction application. Then we propose a new approach for human face recognition based on these moment invariants (CDHMIs and QDHMIs) as descriptors and the Support Vector Machine (SVM) as supervised learning models that analyze data and recognize patterns. Experimental results, obtained using two public datasets, show that the proposed approach is more efficient when the disc-harmonic moments are used instead of other existing descriptors.