<i>c</i>-SNE: Deep Cross-modal Retrieval based on Subjective Information using Stochastic Neighbor Embedding

Cross-modal information retrieval based on subjective aims to enable flexible media services, such as allowing users specify, for example, an image search audio clips. The resulting clips should have impression similar the specified image. Existing methods focus building cross-media cross-modal relationships using objective (such standard caption). However, a relation can be built only between pieces of that are originally related, which limits flexibility retrieval. This research leverages in similarity calculation achieve more flexibility. We propose novel stochastic neighbor embedding technique called c-SNE. c-SNE extract features from and map them common space. It is learning bridge heterogeneous gap modal distributions label-weighted SNE. allows find share same with query medium. Our experimental results benchmark datasets demonstrate proposed method effectively performs distribution alignment Furthermore, our user study ten 600 data points confirmed outperforms three related actual usage situation users' perspective.