RFOVE

RFOVE is an unsupervised, region-based method for approximating an arbitrary 2D shape with a set of ellipses, where both the ellipse parameters and the number of ellipses are determined automatically. It extends the Decremental Ellipse Fitting Algorithm (DEFA) by enabling accurate modeling in cases where the shape is best represented by overlapping ellipses. Like DEFA, RFOVE performs model selection by minimizing the Akaike Information Criterion (AIC) on a shape-complexity measure, but it replaces DEFA’s hard clustering and “Equal Area” constraint with an objective that optimizes shape coverage via the union of shape points covered by all ellipses. This relaxation allows a shape point to contribute to multiple ellipses, improving parameter estimation under overlap and typically achieving higher coverage with fewer ellipses.

RFOVE has been validated on standard datasets, where it consistently yields simpler models (fewer ellipses) with better approximation accuracy than DEFA. As a practical application, RFOVE can be used to detect and segment potentially touching or overlapping cells in fluorescence microscopy images, outperforming prior state-of-the-art methods on public benchmarks, including large datasets containing thousands of annotated cells.