Image Database Modeling

The performance of an indexing technique, namely, a system of feature extraction and similarity measure, is strongly related to the test bed in content-based image retrieval. A theoretical framework is proposed for image database modeling. The goal is to investigate the relationship between a given indexing technique and a given image database. As an analogy to the concept of perplexity of a text corpus, the complexity of an image database is introduced. This is a quantitative measure of the quality of an image database for the task of content access to the database. The complexity of an image database is an internal property of the database and can be used to distinguish the difficulty of different databases on which content access is to be performed. On the other hand, the measure can be used to compare the effectiveness of different indexing techniques with which content access is to be performed.