Edge Detection using Deep Learning

Abstract

“Edge Detection Using Deep Learning”, is a project that is based on research which is key objective to improve a novel detection of edge technique which introduces a pair key flaws of long-term vision recognition by computer issue. We have implemented a deep learning-based edge detection technique termed holistically-nested edge detection in this research. This edge detection introduces two important issues: The first is nested multiscale feature learning, which is encouraged by deep convolutional neural networks for image to image prediction. The second is holistic image training and prediction. An imageto-image estimate is achieved using a HED technique. This method is suggested for determining object boundaries. A deep learning method is used to discover edges that are holistically nested. This model completely combines deeply supervised and fully convolutional neural networks. On the BSD500 dataset and on the NYU Depth dataset where ODS F-score of 0.782 and 0.746 respectively. We do it at a significantly faster pace of 0.4s per image than some recent CNN-based edge detection techniques.