Book contents
- Frontmatter
- Contents
- Preface
- 1 Introduction: Multimedia Applications and Data Management Requirements
- 2 Models for Multimedia Data
- 3 Common Representations of Multimedia Features
- 4 Feature Quality and Independence: Why and How?
- 5 Indexing, Search, and Retrieval of Sequences
- 6 Indexing, Search, and Retrieval of Graphs and Trees
- 7 Indexing, Search, and Retrieval of Vectors
- 8 Clustering Techniques
- 9 Classification
- 10 Ranked Retrieval
- 11 Evaluation of Retrieval
- 12 User Relevance Feedback and Collaborative Filtering
- Bibliography
- Index
- Plate section
4 - Feature Quality and Independence: Why and How?
Published online by Cambridge University Press: 05 July 2014
- Frontmatter
- Contents
- Preface
- 1 Introduction: Multimedia Applications and Data Management Requirements
- 2 Models for Multimedia Data
- 3 Common Representations of Multimedia Features
- 4 Feature Quality and Independence: Why and How?
- 5 Indexing, Search, and Retrieval of Sequences
- 6 Indexing, Search, and Retrieval of Graphs and Trees
- 7 Indexing, Search, and Retrieval of Vectors
- 8 Clustering Techniques
- 9 Classification
- 10 Ranked Retrieval
- 11 Evaluation of Retrieval
- 12 User Relevance Feedback and Collaborative Filtering
- Bibliography
- Index
- Plate section
Summary
For most media types, there are multiple features that one can use for indexing and retrieval. For example, an image can be retrieved based on its color histogram, texture content, or edge distribution, or on the shapes of its segments and their spatial relationships. In fact, even when one considers a single feature type, such as a color histogram, one may be able to choose from multiple alternative sets of base colors to represent images in a given database.
Although it might be argued that storing more features might be better in terms of enabling more ways of accessing the data, in practice indexing more features (or having more feature dimensions to represent the data) is not always an effective way of managing a database:
▪ Naturally, more features extracted mean more storage space, more feature extraction time, and higher cost of index management. In fact, as we see in Chapter 7, some of the index structures require exponential storage space in terms of the features that are used for indexing. Having a large number of features also implies that pairwise object similarity/distance computations will be more expensive.
Although these are valid concerns (for example, storage space and communication bandwidth concerns motivate media compression algorithms), they are not the primary reasons why multimedia databases tend to carefully select the features to be used for indexing and retrieval.
- Type
- Chapter
- Information
- Data Management for Multimedia Retrieval , pp. 143 - 180Publisher: Cambridge University PressPrint publication year: 2010