| Abstract |
In this dissertation, mechanisms for the perceptually-relevant description and retrieval of visual information are presented and discussed. The goal is to derive descriptions of visual information that are compatible with human perception and can be used in queries by content to yield responses that are better appreciated by end users. The proposed mechanisms concern the description of primitive visual features and spatial arrangements of such features, and emphasize the representation of this information with respect to scale of observation. This scale dependent representation is subsequently used to extract image regions with a characteristic spatial arrangement of features and identify gradient-derived dominant structural elements, which are both known to be significant descriptive components of visual content. The perceptual organization of dominant structural elements into perceptual groups yields an additional component of visual content. Attributes of such perceptual groups are then integrated with information about spatial arrangements of primitive visual features and used in the description and retrieval of visual information. In particular, a physiology-inspired method is presented for the representation of primitive visual features based on the scale-summarization of visual content. The scale-summary representation can be computed in parallel and exhibits computational and descriptive properties that extend the standard definition of visual information with respect to scale. In addition, a method is proposed for the multiscale representation of spatial arrangements of primitive visual features. This representation is based on local descriptors and is utilized for the description, classification, and scale-invariant representation of such arrangements. Thus, the information it contains is also used for the browsing and retrieval of visually similar content. Finally, the proposed method is generic, in that it can be used with different local descriptors, and the resulting representation of spatial feature distributions has reduced memory requirements. In this dissertation, mechanisms for the perceptually-relevant description and retrieval of visual information are presented and discussed. The goal is to derive descriptions of visual information that are compatible with human perception and can be used in queries by content to yield responses that are better appreciated by end users. The proposed mechanisms concern the description of primitive visual features and spatial arrangements of such features, and emphasize the representation of this information with respect to scale of observation. This scale dependent representation is subsequently used to extract image regions with a characteristic spatial arrangement of features and identify gradient-derived dominant structural elements, which are both known to be significant descriptive components of visual content. The perceptual organization of dominant structural elements into perceptual groups yields an additional component of visual content. Attributes of such perceptual groups are then integrated with information about spatial arrangements of primitive visual features and used in the description and retrieval of visual information. In particular, a physiology-inspired method is presented for the representation of primitive visual features based on the scale-summarization of visual content. The scale-summary representation can be computed in parallel and exhibits computational and descriptive properties that extend the standard definition of visual information with respect to scale. In addition, a method is proposed for the multiscale representation of spatial arrangements of primitive visual features. This representation is based on local descriptors and is utilized for the description, classification, and scale-invariant representation of such arrangements. Thus, the information it contains is also used for the browsing and retrieval of visually similar content. Finally, the proposed method is generic, in that it can be used with different local descriptors, and the resulting representation of spatial feature distributions has reduced memory requirements. With respect to the component of visual content resulting from the process of perceptual organization, an approach is presented for the description of a certain class of perceptual groups, namely that of image contours, using a curvature-based method that detects perceptually-significant and computationally-stable contour points. These points are then used for the perceptually-relevant piecewise decomposition of contours, which in turn is used for the purpose of contour similarity matching. The integrated use of information about spatial arrangements of primitive visual features and perceptual groups of dominant structure elements is demonstrated and results of representative experiments are presented and discussed. With respect to the component of visual content resulting from the process of perceptual organization, an approach is presented for the description of a certain class of perceptual groups, namely that of image contours, using a curvature-based method that detects perceptually-significant and computationally-stable contour points. These points are then used for the perceptually-relevant piecewise decomposition of contours, which in turn is used for the purpose of contour similarity matching. The integrated use of information about spatial arrangements of primitive visual features and perceptual groups of dominant structure elements is demonstrated and results of representative experiments are presented and discussed.
|
Perceptually Relevant Mechanisms for the Description and Retrieval of Visual Information
