Image retrieval: modelling keywords via low-level features

Abstract

Although Content Based Image Retrieval (CBIR) has attracted large amount of research interest, the difficulties in querying by an example propel ultimate users towards text queries. Searching by text queries yields more effective and accurate results that meet the needs of the users while at the same time preserves their familiarity with the way traditional search engines operate. However, text-based image retrieval requires images to be annotated i.e. they are related to text information. In recent years, much effort has been invested on automatic image annotation methods, since the manual assignment of keywords (which is necessary for text-based image retrieval) is a time consuming and labour intensive procedure. This thesis focuses on image retrieval under the perspective of machine learning and covers di®erent aspects in this area. It discusses and presents several studies referring to: (a) low-level feature extraction and selection for the task of automatic annotation of images, (b) training algorithms that can be utilized for keyword modeling based on visual content, and (c) the creation of appropriate and reliable training data, to be used with the training scheme, using the least manual effort. The main contribution is a new framework that can be used to address the key issues in automatic keyword extraction by creating separate visual models for all available keywords using the one-against-all paradigm to account for the scalability and multiple keyword assignment problems. The prospective reader of this thesis would be equipped with the ability to identify the key issues in automatic image annotation and would be triggered to think ahead to propose alternative solutions. Furthermore, this thesis can serve as a guide for researchers who want to experiment with automatic keyword assignment to digital images.