Author(s)

L.X. Zhou & W.K. Gu

Abstract

This paper presents a novel edge detector based on Feed-Forward Neural Networks (FFNNs). The FFNN computing architecture has two stages, which is a feature enhancement stage as well as a structural boundary extraction stage. The first stage is a traditional supervised BP network, and the second one is manually designed without training. Experiments based on both synthetic and natural images show that the FFNN edge detector can produce accurate, continuous, and smooth edge chains. 1 Introduction Edge detection is one of the most essential problems in computer vision as it is the common starting step for segmentation or feature detection. In natural images an edge is usually difficult to define precisely due to the perceptual component which is associated. In recent years

Keywords