Methods Employed

Introduction

Typically, a CBIR system consists of user interface, query-processing-module and the image database (Figure 1). The user interface is probably the most important part of a search engine. An engine producing perfect results is of no value, if it cannot be handled. It is necessary to reduce the perceived complexity for the user. The approach is to provide graphical guidance while hiding the complex background theories as much as possible. This is especially useful for inexperienced users. Experts have usually far less difficulties in understanding and using complex interfaces. A real life example is the comparison between graphical user interfaces and console applications. Programs started from a console offer several parameters for detailed configuration. This requires knowledge of the program name and the parameter syntax. Their graphical representations are in many cases much easier to handle, but they offer less parameters and actions. For this reason, expert users should always be able to have a lower level access than the beginner.

Figure 1.

Ranking in a typical CBIR-system (Torres et al., 2003)

Browsing

A basic browsing loop during retrieval is shown in Figure 2. The user starts with an initial query, waits for the results and then checks the results. In the best case, the desired content is already displayed on a prominent position on the screen. If not, the user may navigate through more results that are not yet visible on the screen. To improve the results, the initial query may be refined or even rewritten completely to initiate a new search.

Figure 2.

Retrieval-workflow (Pein, Amador, et al., 2008)

This work flow contains all the important stages that need to be presented to the searcher. Each stage requires some methodology and techniques applied in the retrieval system.

For efficient browsing, the user needs to see some content of the available repository. This could be a structure of available categories, the results of a preceding search or other information.

The developed system does only provide a one dimensional list of results, as the aspect of browsing cannot be investigated in depth within this part. This list is always generated by a query which either generates a set of random results or represents a normal retrieval.

Query Language

In section Identification of Problems-Query Language, a set of important aspects for the design of a query language is mentioned. The requirements for the query language applied in this part are derived from these aspects. Converting the users mental model into something machine readable inevitably causes a loss of information. The task of a query language is to minimize this loss. Based on this assumption, the requirements are composed and presented below.