Subspace-based line and curve extraction from noisy images

In this paper, we are interested in the use of subspace tracking techniques for image analysis. More specifically, we present a new model which enables the parametric extraction of lines and curves of any shape. The lines and curves parameters, i.e. angle and offset, are tracked as the image is scanned from the top to the bottom by a sliding window. Based on this model, we propose a new subspace-based adaptive line extraction algorithm (SALE). SALE, an adaptive extension of the SLIDE algorithm, takes advantage of the latest developments in subspace tracking algorithms. SALE outputs a parametric description of detected lines and curves, by tracking their offset and angle throughout the image. Its reduced complexity is directly related to the number of extracted features.