Quantization-free parameter space reduction in ellipse detection

Quantization-free parameter space reduction in ellipse detection Kuang Chung Chen Ellipse modeling and detection is an important task in many computer vision and pattern recognition applications. In this thesis, four Hough-based transform algorithms have been carefully selected, studied and analyzed. These techniques include the Standard Hough Transform, Probabilistic Hough Transform, Randomized Hough Transform and Directional Information for Parameter Space Decomposition. The four algorithms are analyzed and compared against each other in this study using synthetic ellipses. Objects such as noise have been introduced to distract ellipse detection in some of the synthetic ellipse images. To complete the analysis, real world images were used to test each algorithm resulting in the proposal of a new algorithm. The proposed algorithm uses the strengths from each of the analyzed algorithms. This new algorithm uses the same approach as the Directional Information for Parameter Space Decomposition to determine the ellipse center. However, in the process of collecting votes for the ellipse center, pairs of unique edge points voted for the center are also kept in an array. A minimum of two pairs of edge points are required to determine the ellipse. This significantly reduces the usual five dimensional array requirement needed in the Standard Hough Transform. We present results of the experiments with synthetic images demonstrating that the proposed method is more effective and robust to noise. Real world applications on complex real world images are also performed successfully in the experiments.