Deep semantic segmentation-based multiple description coding
- PDF / 2,719,832 Bytes
- 15 Pages / 439.642 x 666.49 pts Page_size
- 52 Downloads / 246 Views
Deep semantic segmentation-based multiple description coding Xue Li1,2 · Lili Meng1,2 Huaxiang Zhang1,2
· Yanyan Tan1,2 · Jia Zhang1,2 · Wenbo Wan1,2 ·
Received: 18 May 2019 / Revised: 7 May 2020 / Accepted: 29 June 2020 / © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract In this paper, we propose a deep semantic segmentation-based multiple description coding (DSSMDC). In the proposed scheme, the input image is divided into two different subsets and getting two descriptions, which is named multiple description pre-processing (MDP). Then, two descriptions are encoded and decoded respectively by utilizing the deep semantic segmentation codec, in which the semantic segmentation label is as side information for improving image reconstruction quality. We can get one side reconstruction, when only one description is received at the decoder. If both descriptions are received at the decoder, we can get the central reconstruction. Experimental results show that the proposed scheme achieves better performance than other existing compression methods. Keywords Multiple description coding · Deep semantic segmentation · Generative adversarial network · Compact network · Reconstruction network
1 Introduction Nowadays, wireless networks and Internet are the main channels of data transmission. There is a rapid development of both new technologies that Internet of Things (IoT) and Cloud Computing (CC) regard the field of wireless communications network [17, 19, 21, 22]. And new transmission algorithm for efficient HEVC delivery can meet the requirements of effective transmission over the Internet [16]. However, the data transmission may result in packet loss, error codes, transmission delay and other problems. Multiple description coding (MDC) is proposed to solve the above problems. For MDC, the source is divided Lili Meng
mengll [email protected] Xue Li [email protected] 1
School of Information Science and Engineering, Shandong Normal University, Jinan, 250014, China
2
Institute of Data Science and Technology, Shandong Normal University, Jinan, 250014, China
Multimedia Tools and Applications
into multiple descriptions, and transmitted through different channels [8]. At the decoder, the more descriptions received, the better the quality of reconstruction image[13]. Multiple description coding as an effective method to solve the problem of packet loss, is firstly produced by Vaishampayan [29]. The multiple description scalar quantizer (MDSQ) method is developed, which is gradually optimal at high rate by using index assignment. In [14], the lattice vector quantization (MDLVQ) design is produced, which is a two-channel multiple description. In [23], a RD-MDC technology is presented, where each JPEG 2000 code-block is still encoded at two rates. In [3], the RD-MDC is extended to M-channel case (M > 2). In [9], the multiple description lapped transform with prediction compensation (MDLTPC) is proposed, where the source signal is divided into two subsets, one subset is the base layer for one des
Data Loading...
