An EMD-SVM screening system for retina digital images: The effect of kernels and parameters

Abstract

The discrete wavelet transform (DWT) and empirical mode decomposition (EMD) are employed to analyze retina digital images in the frequency domain. In particular, statistical features are extracted from high frequency components of the analyzed images. The purpose is to classify normal versus abnormal images. Three different pathologies are considered including, circinates, drusens, and microaneurysms (MA). Support vector machines (SVM) with polynomial and radial basis function kernel are used to classify retina digital images. The simulation results from leave-one-out method (LOOM) show the effectiveness of the EMD-based features over the DWT-based ones. In addition, the polynomial kernel performs better than the radial basis function kernel.