Classification Accuracy with Local and Global Features using LSSVM in Shape Based Image Retrieval Goyal Anjali1, Walia Ekta2, Saini Harvinder Singh3 1Guru Nanak Institute of Management & Technology, Ludhiana (Punjab), India. anjali.garg73@gmail.com 2Faculty, Dept. of Computer Science, South Asian University, New Delhi, India. wekta@yahoo.com 3Guru Nanak Institutions, Ibrahimpatnam (AP), India. hssaini@yahoo.com Online published on 27 June, 2017. Abstract Though Zernike Moments (ZMs) have been most widely used in extracting the region based shape features of an image related to Logo, Trademark, Clip art search etc. Still retrieval of accurate images is an issue. It extracts global features and ignores local details. To overcome this drawback, local features are combined with global features. In this paper, Curvature and Centroid distance are used as local features. These are calculated on the boundary points of edged images. Global features are extracted from ZMs on grayscale image rather than on an edged image. Once features are calculated offline for the whole image database then they are trained with supervised learning method in order to predict the behavior of unseen image. We have used MPEG-7 CE-1 and CE-2 database for the experiments. Least Square Support Vector Machine (LSSVM) is used to implement the learning method which is the fastest one with at par accuracy in comparison to Library for SVM (LIBSVM). The system combined with set of local and global features along with LSSVM contributes towards better classification accuracy. The method of training for LSSVM is one vs one classification in which binary classification is done and is the best way for the retrieval based system. The system is implemented in Matlab 7.4.0. Top Keywords Shape Based Image Retrieval (SBIR), Zernike Moments, local features, Support Vector Machine. Top |