An Euclidean Ellipse Comparison Metric for Quantitative Evaluation
Halil Ibrahim Cakir, Cihan Topal
Accepted in: International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2018.
Ellipse detection is a popular problem in image processing and is utilized in a broad set of image processing applications. Similar to object detection studies, ellipse detection algorithms need to be evaluated quantitatively via the use of datasets. These datasets include ground truth annotations which enable objective assessment to rate the algorithms. However, in contrast to the variety of ellipse datasets in the literature, there is only a few number of ellipse comparison methods to be utilized in matching of annotated and detected ellipses. Moreover, these methods are more like similarity measures and have certain deficiencies which prohibit accurate evaluation of the algorithms. In this study, we propose an ellipse comparison method defined in Euclidean space which accurately compares two ellipses and provides a single quantifiable scalar. Thus, proximity of two ellipses can precisely be estimated without aforementioned flaws and a robust assessment can be performed.