Deep Textual Searching for Visual Semantics of Personal Photo Collections with a Hybrid Similarity Measure

Abstract

In recent years, personal photos on the internet have become important parts of people's lives. There is much automatic application software to search for photos, but searching still suffers from problems of semantic accuracy. Research has been conducted to satisfy user demands for a semantic model using a set of features or keyword annotation techniques. Keywords in photos give the best evidence to identify what photos are about. However, it does not always relate to the actual meaning of photos.For this reason, we propose a textual description with a hierarchical concept and comparison of the feature set with a hybrid similarity measure. The experimental results indicate that our proposed approach offers significant performance improvements in the interpretation of semantic meanings with a maximum success rate of 80.4%.