Cross Media Data Retrieval based on Semantic Consistency

Abstract

Cross media retrieval plays a vital role in finding semantic consistency among the data that are represented through different media. A framework is proposed for this process that starts with Isomorphic Relevant Redundant Transformation (IRRT), which linearly transforms different heterogeneous low-level feature spaces to a high-level redundant feature-isomorphic space without any dimensionality reduction, i.e., without data loss. Then, transforming these data with same dimensionality, with the help of Convex relaxed Alternating Structure Optimization (cASO). Consequently, the SCP for cross-view data can be obtained. On comparing the frameworks using MAP score by keeping Normalized correlation as the distance metric, it is confirmed that the proposed framework forms the better classifier system than the existing framework. So the proposed effective framework has its broad applications in the field of the information retrieval system that works on cross view data.Keywords: Cross-view Data, Cross-Media Retrieval, Semantic Consistency, Semantic Consistent Pattern (SCP), Shared Feature Subspace Learning