Trellis-based circle detection

We introduce a new method for finding circles in images. The proposed method assumes a given pixel is the center of a prospective circle, attempts to fit a circle at that location to the data, and then it scans over all possible pixels. The score at each assumed center location is found by traversing a trellis structure. The trellis allows for gaps in the prospective circles and it enforces a global all or none constraint. The proposed method is compared to a binary matched filter using real visible-spectrum aerial images, where the targets are the circular-shaped silos of surface-to-air missile sites. The proposed method performs noticeably better than the binary matched filter for the data used in this study.