This project investigates new visualization and interaction methods to efficiently present the results of an internet image search engine.
Associate Professor Masahiro Takatsuka.
N/A
A typical image search engine provides its relevance feedback in the form of ranking-based listing or spatial arrangements of searched images. It is contemplated that effective browsing and navigation through the visualized searched images has a potential to significantly improve the searching process. We believe that there are two basic types of information used in defining the similarity measure of digital images; they are attribute features and relational features based on semantic information.
These two types of data form multivariate networks. There are two stages in our approach to visualize this multivariate network. The first stage is dimensional reduction using a Geodesic Self-Organizing Map (GeodesicSOM). The next step is to re-arrange on a circle surrounding the computational element. For each element, the algorithm should try to find a position on the circle that minimizes the number of edge crossings and avoids node to node and node to edge overlap.
The opportunity ID for this research opportunity is 391