A new edge detector based on SMOTE and logistic regression

Abstract

Edge detection remains a hot topic due to its importance as a low level operation for high level operations in computer vision and the fact that there is no edge detector that is optimal for all kinds of images. In this paper, a new edge detector is proposed. The algorithm relies on the concept of edge detection as an imbalanced binary classification problem. In particular, each pixel is characterized by a gradients feature vector and classified as edge or non-edge pixel by means of logistic regression and hysteresis. This algorithm outperforms other state-of-the-art edge detectors both from the visual and quantitative points of view.