Automatic annotation of satellite images with multi class support vector machine

Abstract

Automatic Image Annotation (AIA) is used in image retrieval systems to retrieve the images by predicting tags for images. To achieve image retrieval with high accuracy, an automatic image annotation approach by using Multiclass SVM with the hybrid kernel is proposed. The hybrid kernel is a combination of Radial Basis Function (RBF) and Polynomial Kernel which overcomes the drawbacks of single kernels such as less accuracy, high computational complexity, etc. This technique exploits the Linear Binary Pattern- Discrete Wavelet Transform (LBP-DWT) feature extraction technique to extract the features in horizontal, vertical, and diagonal directions. The experiments suggest that the multiclass SVM can attain a higher accuracy than other conventional SVM with any single kernels. The Multiclass SVM can achieve high accuracy as 95.61% and increases the accuracy by 3.26%, 1.79%, and Kappa coefficient by 3.22%, 2.27% when compared with SVM RBF kernel, polynomial kernel respectively.