Automatic image annotation combining SVMS and KNN algorithm

Abstract

Automatic image annotation has emerged as an important but challenging task in many areas, including web image retrieval, image understanding, Internet data filtering, etc. We consider image annotation task as a Multi-Label Classification (MLC) problem, in which each image can be associated with more than one class. Support Vector Machine (SVMs) performs well in many areas, however it cannot be used to solve the MLC problem directly. While k-Nearest Neighbor algorithm (KNN) has a good performance in MLC problem, an obvious flaw of KNN is its high computational complexity. In this paper, to solve the MLC problem in image annotation with lower computational complexity, we present an image annotation framework combining Support Vector Machine (SVMs) and k-Nearest Neighbor algorithm (KNN). Multiple kinds of features are combined in our method, including edge direction histogram (EDH), gray level co-occurrence matrix (GLCM) and area weighted HSV color histogram. In our system, each image has one label and several tags. Top m possible labels from SVMs result will be as the input of KNN algorithm. Experiments conducted on the typical Corel dataset shows the scheme has obtained higher accuracy.