ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) 2020
DOI: 10.1109/icassp40776.2020.9054078
|View full text |Cite
|
Sign up to set email alerts
|

Deep Joint Source-Channel Coding for Wireless Image Retrieval

Abstract: Motivated by surveillance applications with wireless cameras or drones, we consider the problem of image retrieval over a wireless channel. Conventional systems apply lossy compression on query images to reduce the data that must be transmitted over the bandwidth and power limited wireless link. We first note that reconstructing the original image is not needed for retrieval tasks; hence, we introduce a deep neutral network (DNN) based compression scheme targeting the retrieval task. Then, we completely remove… Show more

Help me understand this report
View preprint versions

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
2
1

Citation Types

0
17
0

Year Published

2020
2020
2024
2024

Publication Types

Select...
4
3
1

Relationship

1
7

Authors

Journals

citations
Cited by 44 publications
(17 citation statements)
references
References 10 publications
0
17
0
Order By: Relevance
“…Along this line, Yang et al [25] further compressed the transmitted data by introducing the semantic relationship between feature maps and semantic concept. Jankowski et al [26] proposed a task-based compression scheme for input images for the person re-ID task, and pointed that in classification it is unnecessary to send feature maps to the receiver because the transmitter can execute the task locally and send only the predictive class label, in contrast, the transmitter can't perform image retrieval alone due to missing gallery images.…”
Section: B Framework Of Semantic Communication Systemsmentioning
confidence: 99%
See 1 more Smart Citation
“…Along this line, Yang et al [25] further compressed the transmitted data by introducing the semantic relationship between feature maps and semantic concept. Jankowski et al [26] proposed a task-based compression scheme for input images for the person re-ID task, and pointed that in classification it is unnecessary to send feature maps to the receiver because the transmitter can execute the task locally and send only the predictive class label, in contrast, the transmitter can't perform image retrieval alone due to missing gallery images.…”
Section: B Framework Of Semantic Communication Systemsmentioning
confidence: 99%
“…• Based on the traditional rate-distortion theory and information bottleneck principle, an extended rate-distortion theory is proposed, which aims to optimized the communication system for the purpose of trade-off between concise representation and semantic distortion, where the semantic distortion is referred to the proper trade-off [24]- [26], (c) recently proposed multitask type semantic communication with reconstruction [4], [27]- [31], and (d) our proposed semantic communication system with reconstruction. Note that the channel codding and decoding are ignored, which is not we focus on.…”
Section: Introduction Semantic Communication Is Categorized By Shanno...mentioning
confidence: 99%
“…A practical edge inference problem is studied in [4], where the image of a person captured by a remote camera is to be identified within a database available at an edge server, called the reidentification (re-ID) problem. Here, the camera cannot make a local decision as it does not have access to the database.…”
Section: Distributed Inferencementioning
confidence: 99%
“…Here, the camera cannot make a local decision as it does not have access to the database. In [4], two approaches are proposed, both employing DNNs for remote inference: a task-oriented DNN-based compression scheme for digital transmission and a DNNbased analog JSCC approach,à la DeepJSCC. These schemes are compared in Fig.…”
Section: Distributed Inferencementioning
confidence: 99%
“…In [15] and [16], a CNN based AE is used to compress and transmit images. In [17]- [19], AE is used to compress the intermediate feature of a STL network. These deep models are trained with noise, so they are robust to channel interference.…”
Section: Introductionmentioning
confidence: 99%