| Two-dimensional shape recognition has become a prevalent research area in machine vision as the amount of data in the world increases each year. Shape recognition is used in such diverse areas as robotics, manufacturing, quality control, psychology, biology, and medicine as a tool for both feature detection and image classification. While there are numerous methodologies presented in the literature for shape recognition, very few have the capability to determine the scale and orientation of a matched shape to its library template while maintaining a sparse library and robustness to noise. This thesis presents a two dimensional shape recognition method using sparse complex Fourier feature sets that are robust to scale, orientation, and noise. A method for determining the scale and orientation of the matched object using the proposed feature sets is also discussed. Finally, the method is extended to the application of three dimensional shape recognition through the use of two-dimensional slicing and principal component analysis.;The main contibution of this thesis to current shape recognition algorithms which use the Fourier transform for their feature basis is the exploitation of the phase information for rotation, magnitude information for scale, and sparsity measurements that can be used to reduce the feature set size. |
