A Probabilistic Mixture Approach to Automatic Ellipse Detection

Ellipse detection from a digital image, especially with a complicated background, is still a very challenging problem in image analysis and understanding, and its difficulty relies on how to effectively model these ellipses from edge pixels and locate them automatically. In this paper, we propose a probabilistic mixture model for the probable ellipses in the image and implement a Bayesian Ying-Yang (BYY) harmony learning algorithm to learn the parameters of the mixture with automatic model selection so that the correct number of ellipses can be detected and located in an automatic way. Various simulation experiments demonstrate better and robust performance by compared with the current state-of-the-art ellipse detection methods. Moreover, it is successfully applied to the elliptical shape detection in complex real-world images and the fetal abdominal segmentation in ultrasound images.