Multi-frequency transformation for edge detection

Wenlong Fu, Mark Johnston, Mengjie Zhang

Abstract: This paper investigates the use of a multi-frequency transformation for edge detection. Most edge detectors are good at detecting non-texture edges, but have problems with texture edges. In order to detect texture edges, prior knowledge is usually required to avoid heavy computational cost. In this study, a fast and simple transformation based on multi-frequencies is proposed to improve detection performance and the relevant analysis for proper responses on texture and non-texture edges is given. The experimental results show that a classical edge detector improves detection performance after using the proposed transformation based on multi-frequencies, and the detection result from the edge detector using the transformation is better than the detection result from some popular feature extraction techniques, such as extraction based on Gaussian gradients, histogram gradients, and surround suppression.