Bridging the semantic gap for automatic image annotation by learning the manifold space

Automatic image annotation has been an active research topic in recent years due to its potential impact on image understanding and web image mining. In this regard, image feature vectors consist of low level features like color, texture, shape and object spatial relations. But in many situations the similarity between two images could not be found correctly by the Euclidean distance. The aim of basic non-linear methods in feature extraction is to find the intrinsic dimensions where each dimension indicates a latent feature. The purpose of this study is to reduce the dimensions of feature vectors by a non-linear approach, named manifold learning, and develop a new feature vector to coincide semantic and Euclidean distance. So, the continuity between the instances of a semantic at the semantic space is kept in feature space. Keeping the continuity at feature space is the main approach to decrease the semantic gap in this study. The experiments showed that the geometrical distances between the samples in this approach are closer to their semantic distance. The proposed method has been compared to the other well-known approaches on Corel and IAPR-TC12 datasets. The results confirmed the effectiveness and validity of the proposed method.