| Abstract |
In recent years, images are being generated at an ever-increasing rate in application areas such as home entertainment systems, medicine, remote sensing etc. The great volume, variety and special characteristics of image data require that methods and systems for the efficient management of such data be developed. In order to store, organize and retrieve such large amounts of information specialized Image Database Management Systems have been created. The main disadvantage of these systems is that image retrieval is based on alphanumeric data. A small number of such systems support the retrieval of relevant images based exclusively on their pictorial content. Our goal is to embody the biological model of vision in the techniques of machine vision in order to extract characteristic components of image content. The laws of Gestalt Theory, from the discipline of Psychology, are used for this purpose. This theory, which was first announced at the beginning of our century, reasons out our ability to impose structural organizations on sensory data, so as to group sensory primitives arising from a common underlying cause. The representative law of this theory is the law of Pragnanz. This law simply states that perceptual groupings tend to be ``good gestalts'' or ``good figures'' and that this organization has the properties of regularity, simplicity and stability over time. The laws of perceptual organization are used in order to detect perceptually significant groups. These groups are formed by primitive image features, such as linear segments. The laws of perceptual organization enable us to detect the groups that are most probable to have come from the same objects, without us having an a priori knowledge of the objects contained in the scene. The relations of proximity, parallelism, collinearity, continuity and vertical direction are used for this purpose. A bottom up hierarchical process is used to detect associations between features that satisfy the above relations. Salient perceptual groups are generated by the associations that are most probable to appear. The grouping results are organized hierarchically.
|
Ομαδοποίηση χαρακτηριστικών των ψηφιακών εικόνων με βάση τις αρχές της αντιληπτικής οργάνωσης
