LAUSR

Abstract Active contour model (ACM) is able to obtain sub-pixel precision segmentation and has been widely employed in biomedical image analysis and video segmentation. Existing region-based ACMs (RACM) without enough prior constraints, however, easily fail when segmenting low quality images, e.g. biomedical images corrupted by strong noise and intensity inhomogeneity simultaneously. In this paper, we propose frequency boundary energy (FBE), a generalized RACM, thus provide a new perspective to understand RACM, whose segmentation results are determined by some predefined frequency filters. We then introduce difference of Gaussians (DoG) as a better filter to exclude strong noise and intensity inhomogeneity effectively for RACM. We show that this new model, namely FBE–DoG, possesses three major advantages, i.e. allowing selective segmentation, with controllable smoothness and being able to segment near-regular texture images without using complex texture feature. We compare FBE–DoG with state-of-the-art RACM methods and show its superior performance on challenging biomedical image dataset and a real-world optical coherence tomography image sequence.