Multi-Scale Detection of Curvilinear Structures with High Contour Accuracy

The detection of lines is an important low-level operation in computer vision that has many applications. Most of the existing line detectors do not take into account the width of the lines to be detected. In this paper, a novel multi-scale approach to curvilinear features detection with high contour accuracy is presented. A line detector at different orientations is applied to a multi-scale decomposition of the original image in order to obtain scale-orientation signatures at each pixel, which can be used to detect seeds. The contours of the linear structures are fitted by growing these seeds. The performance of the algorithm is evaluated by its application to recognize marbling in Iberian ham images.