Spatial Pyramid Based 3D Hog Feature Extraction for Alzheimer’s Disease Identification

Abstract

Image based machine learning is widely used for Alzheimer’s disease (AD) identification, however, traditional two dimensional (2D) image processing techniques may lead to spatial information lost. In this paper, a spatial pyramid based three dimensional (3D) histogram of oriented gradient (HOG) feature extraction method is proposed to increase spatial discrimination. Then a simple f-divergence based dimension reduction method was used to extract principal features in HOG, which solved the problem of high dimensionality of variables with relatively small number of subjects. Experimental results illustrated the effectiveness of proposed method.