Field of Junctions: Extracting Boundary Structure at Low SNR
This work addresses the problem of robust boundary detection and segmentation for computer vision systems operating in noisy environments, offering a unified approach to handle different boundary elements.
This paper introduces a bottom-up model, the 'field of junctions', to simultaneously detect various boundary elements like contours, corners, and junctions in images. The model uses a 'generalized M-junction' to explain boundary shape in image patches and employs non-convex optimization with a novel regularizer to enforce spatial consistency, allowing it to succeed even at high noise levels where other methods fail.
We introduce a bottom-up model for simultaneously finding many boundary elements in an image, including contours, corners and junctions. The model explains boundary shape in each small patch using a 'generalized M-junction' comprising M angles and a freely-moving vertex. Images are analyzed using non-convex optimization to cooperatively find M+2 junction values at every location, with spatial consistency being enforced by a novel regularizer that reduces curvature while preserving corners and junctions. The resulting 'field of junctions' is simultaneously a contour detector, corner/junction detector, and boundary-aware smoothing of regional appearance. Notably, its unified analysis of contours, corners, junctions and uniform regions allows it to succeed at high noise levels, where other methods for segmentation and boundary detection fail.