Paper ID | IVMSP-11.5 | ||
Paper Title | Decouple the High-Frequency and Low-Frequency Information of Images for Semantic Segmentation | ||
Authors | Lianlei Shan, Xiaobin Li, Weiqiang Wang, University of Chinese Academy of Sciences, China | ||
Session | IVMSP-11: Image & Video Segmentation | ||
Location | Gather.Town | ||
Session Time: | Wednesday, 09 June, 14:00 - 14:45 | ||
Presentation Time: | Wednesday, 09 June, 14:00 - 14:45 | ||
Presentation | Poster | ||
Topic | Image, Video, and Multidimensional Signal Processing: [IVCOM] Image & Video Communications | ||
IEEE Xplore Open Preview | Click here to view in IEEE Xplore | ||
Abstract | As a special kind of signal processing technology, image processing has been developed rapidly after the appearance of convolutional neural network (CNN). However, until now, CNN has not been well integrated with the traditional image processing methods, which makes the former traditional image processing methods lost their meaning. In this paper, we extract the high and low frequency components of images via the classical Fourier transform and high-pass and low-pass filters in digital signal processing. Based on one reasonable empirical assumption that the high-frequency component mostly represents the edge part and the low-frequency component represents the body part, high and low frequency components are put into one two branch network to extract the body and the edge information respectively. In semantic segmentation, the two main developing trends are to maintain the consistency of the internal features of same categories and to supervise the edge more clearly. And our proposed approach both ensures consistency of the internal features and enhances edge supervision in one single network, and results on Cityscapes and KITTI observably show the effectiveness of our work. |