Daniel Blank

Clustering-Based Source Selection for Efficient Image Retrieval in Peer-to-Peer Networks

In peer-to-peer (P2P) networks, computers with equal rights form a logical (overlay) network in order to provide a common service that lies beyond the capacity of every single participant. Efficient similarity search is generally recognized as a frontier in research about P2P systems. One way to address it is using data source selection based approaches where peers summarize the data they contribute to the network, generating typically one summary per peer. When processing queries, these summaries are used to choose the peers (data sources) that are most likely to contribute to the query result. Only those data sources are contacted. There are two main contributions of this paper. We extend earlier work, adding a data source selection method for high-dimensional vector data, comparing different peer ranking schemes. More importantly, we present a method that uses progressive stepwise data exchange between peers to better each peers summary and therefore improve the systems performance

A hybrid index structure for geo-textual searches

The efficient execution of multi-criteria queries has gained increasing interest over the last years. In the present paper we propose an R-tree based approach for queries addressing textual as well as geographic filter conditions. Whereas most previous approaches use an index structure optimised for a single criterion adding special treatment for the other criterion at the leaf nodes or end points of this index structure, our approach uses a deeper integration. In short, R-trees are maintained for certain subsets of the whole term set. Furthermore, in each of these R-trees bit sets are used within the nodes to indicate whether entries for the terms associated with the single bits can be found in the corresponding sub-tree. Our index structure aims to be both, time and space efficient. The paper investigates the efficiency and applicability of the proposed index structure via practical experiments based on real-world and synthetic data.

Sample-based creation of peer summaries for efficient similarity search in scalable peer-to-peer networks

In this paper we introduce a simple yet experimentally convincing approach in the research field of source selection for content-based similarity search in P2P networks or, more concretely, in summary-based P2P systems. In these systems, summaries are used for data source selection when performing k-NN queries on distributed collections of documents represented by feature vectors. We introduce a new type of cluster-based summaries for source selection that can efficiently and cheaply be calculated and distributed in P2P networks. For the summaries generation, a very large number of sample points is used. Each peer in the network assigns its indexing data to their corresponding closest sample points and publishes its constructed summary. We evaluate the quality of these summaries when changing the number of sample points used in experiments on real-world image feature data obtained from a large crawl of the flickr web photo community and show that for higher numbers of sample points we achieve a better retrieval performance. Our experiments show that the proposed summaries yield four times better performance with respect to previous methods. Intuitively, there are some disadvantages to this approach due to the large size of the generated summaries. We show experimentally, that these disadvantages can easily be overcome due to the sparse nature of the generated summaries by simple compression techniques.

Picadomo: Faceted Image Browsing for Mobile Devices

Picadomo combines content-based image retrieval and faceted search on mobile devices. It is designed for finding images with desired visual properties, tags or other known metadata. Due to the limitations of mobile devices such as small screen sizes and low processing power, we had to carefully select the features that come in use (dominant color, GPS data, tags, etc.). With Picadomo the user can pick visualized facets directly via touch screen, while using very little screen size for the facet browsing navigation. We present our architecture, the facets used for image browsing, our new control concept and user experiments.

A depth-first branch-and-bound algorithm for geocoding historic itinerary tables

The work in this paper is motivated from two different perspectives: First, gazetteers as an important data source for Geographic Information Retrieval (GIR) applications often lack historic place name information. More focused historic gazetteers are a far cry from being complete and often specialize only on certain geographic regions or time periods. Second, research on historic route descriptions---so called itineraries---is an important task in many research disciplines such as geography, linguistics, history, religion, or even medicine. This research on historic itineraries is characterized by manual, time-consuming work with only minimalistic IT support through gazetteers and map services. We address both perspectives and present a depth-first branch-and-bound (DFBnB) algorithm for deducing historic place names and thus the stops of ancient travel routes from itinerary tables. Multiple phonetic and character-based string distances are evaluated when resolving parts of an itinerary first published in 1563.

Geocoding place names from historic route descriptions

Many gazetteers contain only a small amount of historic place name information and spelling variants of places. Even more focused historic gazetteers are far from being complete and often specialize on certain geographic regions or particular time periods. On the other hand, there are huge amounts of historic route descriptions, so called itineraries. They represent massive knowledge sources from which historic place names and spelling variants can be deduced. The analysis and geocoding of those route descriptions---often done by hand---is an important task in the humanities. To cope with these problems, we present preliminary ideas how to automatically deduce historic place names and thus travel routes from historic route descriptions.

Using Hybrid Techniques for Resource Description and Selection in the Context of Distributed Geographic Information Retrieval

The amount of media items on the web is increasing tremendously, especially regarding personal media items. To effectively collaborate over and share these massive amounts of media objects, there is a strong need for adequate indexing and search techniques. Trends like social networks, large-storage mobile devices and high-bandwidth networks make peer-to-peer (P2P) information retrieval systems of deep interest. Hence, resource selection based on compact resource descriptions is used to efficiently determine promising peers w.r.t. a query. To design effective media search applications, multiple search criteria need to be addressed. Subsequently, besides text or visual media content, geospatial data is frequently used. We propose techniques to summarize and select collections of georeferenced media items in P2P systems. Generally, these summarization techniques can be divided into geometric and space partitioning approaches. This paper presents and evaluates techniques of a third category, hybrid approaches that combine features of geometric and space partitioning techniques.

Visualflamenco: Dependable, interactive image browsing based on visual properties

Faceted search as a method for exploratory search is a search interface technique for Boolean retrieval that provides a good compromise between the wish to guide the user and the need to give freedom to the user. Both, guidance and freedom are needed in order to enable him (or her) to accomplish his search goal as quickly and efficiently as possible. The search goal can be both, to retrieve a target image (lookup search) or to get an overview of a collection as the basis for accomplishing complex search tasks (exploratory search). VisualFlamenco is about bringing content-based image retrieval and faceted search together. While there is some preexisting work towards faceted search based on automatically extracted metadata, the base innovation of our work is to visualize the meaning of facets and let the user pick visualized facets directly. The main point is this research is to give the user the right expectations about what the system can achieve. We see this as a step towards dependable multimedia search, search interfaces that do not promise more than they can achieve. A mismatch of expectations and possibilities has been denoted by some as one of the main problems in current research about multimedia search. Within this paper we present a visual faceted search tool that seeks to provide dependable image search. We describe its base idea, the features used, as well as user experiments done using the system.

CLUSTERING-BASED, LOAD BALANCED SOURCE SELECTION FOR CBIR IN P2P NETWORKS

In peer-to-peer (P2P) networks, computers with equal rights form a logical (overlay) network in order to provide a common service that lies beyond the capacity of every single participant. Efficient similarity search is generally recognized as a frontier in research about P2P systems. One way to address this issue is using data source selection based approaches where peers summarize the data they contribute to the network, generating typically one summary per peer. When processing queries, these summaries are used to choose the peers (data sources) that are most likely to contribute to the query result. Only those data sources are contacted. There are several contributions of this article. We extend earlier work, adding a data source selection method for high-dimensional vector data, comparing different peer ranking schemes. Furthermore, we present two methods that use progressive stepwise data exchange between peers to better each peers summary and therefore improve the systems performance. We finally examine the effect of these data exchange methods with respect to load balancing.

Using Summaries to Search and Visualize Distributed Resources Addressing Spatial and Multimedia Features

Summarization is an important means to cope with the challenges of big data. Summaries can help to achieve a first overview, they can be used to characterize subsets, they allow for the targeted access to data, and they build the basis for visualization techniques. In the present article, we point out the role of summaries as well as potential application scenarios. As examples, summarization techniques for spatial data (as an example for specific low dimensional techniques) and for general metric spaces (as a generic example with a broad spectrum of applications) are described. Furthermore, their use for resource selection and resource visualization in large distributed scenarios is outlined.