Browse Theses

The objective of this research is to develop a new knowledge-based image description system using fractal features. We incorporate a knowledge-based recognition process into this image description system to provide us with the capability of natural scene understanding. In this dissertation, the structure of this system and its interaction strategy for the extraction of scene description are also presented with experimental results.

This research is divided into three phases: fractal feature development, image segmentation with fractal features, and knowledge-based scene description. First, a novel fractal model is developed to efficiently extract fractal features from the natural image. We use fractal features to represent the texture characteristics of natural surfaces. Second, we develop a image segmentation algorithm which can segment natural images into regions using the developed fractal features. We also conduct a number of texture classification experiments in this phase to demonstrate the capability of these fractal features. Third, a knowledge-based scene recognition process is used to describe natural images through the interaction between the knowledge base and the system. An extensive survey of previous work in this area and its relative advantages/disadvantages has been included in this dissertation.

Three sets of gray level images are used in the experiment. They are synthesized fractal images, natural textures, and natural scene images. The first two sets of images are used to develop the fractal model and study its characteristics. Based on the developed fractal model, the sets of scene images are used in segmentation and scene recognition processes.