Contextual edge detection using a recurrent neural network

Abstract

If we consider edge detection as a classification problem, then it seems reasonable that context should play an important role in its study. In fact, it is frequent that neighboring pixels exhibit a strong inter-dependence. In this paper we propose a recurrent neural network for edge detection, which uses a special architecture intended to incorporate contextual information during operation. Some experimental results are presented, showing its effectiveness.