Image databases : using perceptual organization, color and texture for retrieval in digital libraries

The focus of this research is to develop a comprehensive framework for contentbased retrieval in digital image libraries. Most of the current techniques in image retrieval are oriented towards lower-level processing of image data (pixel color histogram and image texture analysis). These techniques rarely analyze an image for the extraction of higher-level semantic features that describe the structural content of an image. Perceptual organization, grouping and inference principles are used in this work for the extraction of semantic information exhibited in the form of image structure. The usefulness of higher-level features for the retrieval of images containing manmade objects, such as buildings, towers, bridges, and architectural objects, is studied and demonstrated. Their utility is also exploited in a general framework for image retrieval, where queries are served by a retrieval methodology composed of both lower-level and higher-level analyses. The synergy resulting from this combination helps in successfully retrieving a wide variety of images, including those containing natural objects such as trees, water, sky, mountains, animals, and landscapes. In addition, this dissertation presents perceptual constancy in grouping and organization as the link between perception, geometry and transformation. The perceptual organizational phenomenon of shape constancy is used as a key idea in representing structural similarity. A unified representation of structural similarity as a concept of invariance over the group of Euclidean similarity transformations is presented. A similarity-invariant model of perceptual organization and grouping is derived, and its applications are considered. The retrieval process is also examined using a mathematical model of isotropic and anisotropic mappings. Users in many fields are exploiting the opportunities offered by the ability to access and manipulate collections of digital images. They include people from diverse disciplines, such as medicine, architecture, engineering, fashion, graphic design, publishing, crime prevention, as well as ordinary users on the internet. The developed system has applications in the manipulation and organization of images in digital archives and the World-Wide-Web (WWW), serving the needs of a large number of users.