Adversarial Network for edge detection
Refereed conference paper presented and published in conference proceedings


摘要Edge detection is a fundamental problem in computer vision and has been explored for many decades. Due to the rapid development of machine learning techniques and their applications to image processing, there is a proliferation of neural network-based approaches to solve the edge detection problem. These methods have good performance and even outperform human beings. Most of the existing neural networkbased systems use the convolutional network or its variant. They usually produce thick edges and the application of nonmaximum-suppression to suppress the edge is necessary. In this paper, we explore another type of neural network called the conditional generative adversarial network (cGAN) to address the edge detection problem. cGAN is an innovative framework to do the image synthesis task. It can generate an
mage close to the real one. After training, our network can produce an edge map that contains more detailed information and thinner edges compared to the state-of-the-art methods that require the well-known non-maximum-suppression for post-processing. The proposed approach is able to produce a high quality edge map directly without further processing. Our solution is computation efficient. It can achieve a speed of 59 and 26 frames per second (fps) for an image resolution of 256x256 and 512x512, respectively.
著者Zhiliang Zeng, Ying Kin Yu, Kin Hong Wong
會議名稱2018 Joint 7th International Conference on Informatics, Electronics & Vision (ICIEV) and 2018 2nd International Conference on Imaging, Vision & Pattern Recognition (icIVPR)
頁次19 - 23

上次更新時間 2018-21-11 於 10:39