Shape and Image Analysis using Neural Networks Fractals and Wavelets
Abstract
A general theory and appropriate statistical analysis are developed for discrimination of objects by shape, i.e., by using features which are invariant to location, scale and orientation, in particular to reflection also. In the case of landmark data, features such as Euclidean distances between landmarks or angles of triangles after suitable triangulation are considered. When the objects do not have recognizable landmarks, as in the case of closed boundaries, we can use topological properties of points on the boundary at intervals of constant length as features. To deal with such cases, a new geometry of circular vectors with a suitably defined metric is developed. This enables the use of distance methods such as k-NN rule in pattern recognition. We have also concentrated on the extraction of features for representing shapes. As a generalization we have considered the signals from machining process and studied characterization using chaos and fractal analysis. We extended this work to represent shapes using wavelets, Fourier descriptors, fractal image compression and iterated functional systems. We have conducted a comparative analysis. In the contemporary internet world search engines need sophisticated techniques to search for images of interest based on shapes. We proposed a preliminary model and web bot based upon our shape analysis study to develop an image search engine for the www.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 07, 2000
- Accession Number
- ADA388833
Entities
People
- Calyampudi Radhakrishna Rao
- S. R. Kumara
Organizations
- Pennsylvania State University