Siwang Zhou
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Siwang Zhou has a total of 65 co-authors. Here are the top co-authors by number of shared publications:
- Yonghe Liu
Department of Computer Science and Engineering, The University of Texas at Arlington, Arlington, TX 76019, USA
8 shared publications - Hongbo Jiang
Hunan University
7 shared publications - Wei Zhang
School of Computer Science and Engineering, Changsha University, Changsha, China
7 shared publications - Keqin Li
Hunan University
5 shared publications - Xingting Liu
College of Computer Science and Electrical Engineering, Hunan University, Changsha, China
5 shared publications - Daibo Liu
College of Computer Science and Electronic Engineering, Hunan University, Changsha, China
4 shared publications
Latest Publications
- 13 November 2025
- Published by Elsevier in Journal of Visual Communication and Image Representation
- Vol. 114
Facial image super-resolution network for confusing arbitrary gender classifiers
- 13 November 2025
- Published by Elsevier in Journal of Visual Communication and Image Representation
- Vol. 114
Facial image super-resolution network for confusing arbitrary gender classifiers
- Open Access
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 3114-3123
Mitigating Delivery Artifacts in Real-World Video Super-Resolution
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 3114-3123
Open AccessMitigating Delivery Artifacts in Real-World Video Super-Resolution
- Open Access
- 11 April 2025
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 39 (5) , 4761-4769
Block-Based Multi-Scale Image Rescaling
Image rescaling (IR) seeks to determine the optimal low-resolution (LR) representation of a high-resolution (HR) image to reconstruct a high-quality super-resolution (SR) image. Typically, HR images with resolutions exceeding 2K possess rich information that is unevenly distributed across the image. Traditional image rescaling methods often fall short because they focus solely on the overall scaling rate, ignoring the varying amounts of information in different parts of the image. To address this limitation, we propose a Block-Based Multi-Scale Image Rescaling Framework (BBMR), tailored for IR tasks involving HR images of 2K resolution and higher. BBMR consists of two main components: the Downscaling Module and the Upscaling Module. In the Downscaling Module, the HR image is segmented into sub-blocks of equal size, with each sub-block receiving a dynamically allocated scaling rate while maintaining a constant overall scaling rate. For the Upscaling Module, we introduce the Joint Super-Resolution method (JointSR), which performs SR on these sub-blocks with varying scaling rates and effectively eliminates blocking artifacts. Experimental results demonstrate that BBMR significantly enhances the SR image quality in the of 2K and 4K test dataset compared to initial network image rescaling methods.
- 11 April 2025
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 39 (5) , 4761-4769
Open AccessBlock-Based Multi-Scale Image Rescaling
Image rescaling (IR) seeks to determine the optimal low-resolution (LR) representation of a high-resolution (HR) image to reconstruct a high-quality super-resolution (SR) image. Typically, HR images with resolutions exceeding 2K possess rich information that is unevenly distributed across the image. Traditional image rescaling methods often fall short because they focus solely on the overall scaling rate, ignoring the varying amounts of information in different parts of the image. To address this limitation, we propose a Block-Based Multi-Scale Image Rescaling Framework (BBMR), tailored for IR tasks involving HR images of 2K resolution and higher. BBMR consists of two main components: the Downscaling Module and the Upscaling Module. In the Downscaling Module, the HR image is segmented into sub-blocks of equal size, with each sub-block receiving a dynamically allocated scaling rate while maintaining a constant overall scaling rate. For the Upscaling Module, we introduce the Joint Super-Resolution method (JointSR), which performs SR on these sub-blocks with varying scaling rates and effectively eliminates blocking artifacts. Experimental results demonstrate that BBMR significantly enhances the SR image quality in the of 2K and 4K test dataset compared to initial network image rescaling methods.
- 24 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Broadcasting
- Vol. 71 (2) , 641-652
Parameter Estimation for Adaptive Impulsive Noise Suppression: A Deep Learning-Based Memoryless Nonlinearity Approach
In the OFDM-based digital terrestrial broadcasting systems, impulsive noise is a significant factor affecting communication quality. A prominent method to suppress impulsive noise is to incorporate a memoryless nonlinearity at the receiver front-end of the OFDM demodulator, in which parameter estimation of memoryless nonlinearity directly impact the effectiveness of impulsive noise suppression. In this paper, we proposes a deep learning-based memoryless nonlinearity approach for impulsive noise suppression. The proposed method can adaptively estimate the parameters of the memoryless nonlinearity in dynamic impulsive noise environments and achieve totically-optimal parameter estimation. To specific, we design a High-Amplitude Priority Downsampling method to extract the key amplitude characteristics from the input signal, which effectively resolves the issue of extracting amplitude features of impulsive noise. Besides, to address the issue of performance degradation due to insufficient training samples, we propose a novel training method that integrates progressive fine-tuning to complete the training only using few samples. Furthermore, we conduct experiments on signal-to-noise ratio (SNR) and bit error rate (BER) of the signal after impulsive noise suppression. The results validate that the parameters estimated by the proposed method can approximate the theoretical optimal values and the proposed method can effectively suppress impulsive noise and outperform the traditional methods in terms of SNR and BER.
- 24 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Broadcasting
- Vol. 71 (2) , 641-652
Parameter Estimation for Adaptive Impulsive Noise Suppression: A Deep Learning-Based Memoryless Nonlinearity Approach
In the OFDM-based digital terrestrial broadcasting systems, impulsive noise is a significant factor affecting communication quality. A prominent method to suppress impulsive noise is to incorporate a memoryless nonlinearity at the receiver front-end of the OFDM demodulator, in which parameter estimation of memoryless nonlinearity directly impact the effectiveness of impulsive noise suppression. In this paper, we proposes a deep learning-based memoryless nonlinearity approach for impulsive noise suppression. The proposed method can adaptively estimate the parameters of the memoryless nonlinearity in dynamic impulsive noise environments and achieve totically-optimal parameter estimation. To specific, we design a High-Amplitude Priority Downsampling method to extract the key amplitude characteristics from the input signal, which effectively resolves the issue of extracting amplitude features of impulsive noise. Besides, to address the issue of performance degradation due to insufficient training samples, we propose a novel training method that integrates progressive fine-tuning to complete the training only using few samples. Furthermore, we conduct experiments on signal-to-noise ratio (SNR) and bit error rate (BER) of the signal after impulsive noise suppression. The results validate that the parameters estimated by the proposed method can approximate the theoretical optimal values and the proposed method can effectively suppress impulsive noise and outperform the traditional methods in terms of SNR and BER.
- Open Access
- 7 March 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Multimedia Computing, Communications, and Applications
- Vol. 21 (3) , 1-22
GFPNet: Generalizable Face Privacy Network with Dynamic Defense Training
In certain specific scenarios, there is a risk of privacy leakage in terms of the soft biometric attributes on a person’s face. However, existing face privacy-enhancing techniques suffer from limited generalizability, meaning that they can only induce misclassification in a specific classifier but fail to generalize this effect well to arbitrary attribute classifiers. Moreover, existing methods reverse attributes to improve face privacy, but this may result in privacy recovery. To address those problems, we propose GFPNet, a novel privacy-enhancing model that can provide generalizable and reliable privacy to face images. The key factor for improving generalizability is that GFPNet uses defense training, which is an effective way to improve model robustness, to dynamically strengthen the mediocre auxiliary attribute classifier during iterative training. Specifically, the generalizability of GFPNet is enhanced in the game between attack and defense, where the generator attempts to deceive the auxiliary attribute classifier and the classifier defends against the generator’s attack by defense training. Furthermore, instead of reversing attributes, skewing attributes to one side is used to avoid attribute recovery. GFPNet also integrates a face matcher, multi-scale discriminator, and Demiguise Attack to improve face matching and image quality. Extensive experiments demonstrate that GFPNet has excellent generalizability to arbitrary attribute classifiers and satisfactory face-matching utility.
- 7 March 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Multimedia Computing, Communications, and Applications
- Vol. 21 (3) , 1-22
Open AccessGFPNet: Generalizable Face Privacy Network with Dynamic Defense Training
In certain specific scenarios, there is a risk of privacy leakage in terms of the soft biometric attributes on a person’s face. However, existing face privacy-enhancing techniques suffer from limited generalizability, meaning that they can only induce misclassification in a specific classifier but fail to generalize this effect well to arbitrary attribute classifiers. Moreover, existing methods reverse attributes to improve face privacy, but this may result in privacy recovery. To address those problems, we propose GFPNet, a novel privacy-enhancing model that can provide generalizable and reliable privacy to face images. The key factor for improving generalizability is that GFPNet uses defense training, which is an effective way to improve model robustness, to dynamically strengthen the mediocre auxiliary attribute classifier during iterative training. Specifically, the generalizability of GFPNet is enhanced in the game between attack and defense, where the generator attempts to deceive the auxiliary attribute classifier and the classifier defends against the generator’s attack by defense training. Furthermore, instead of reversing attributes, skewing attributes to one side is used to avoid attribute recovery. GFPNet also integrates a face matcher, multi-scale discriminator, and Demiguise Attack to improve face matching and image quality. Extensive experiments demonstrate that GFPNet has excellent generalizability to arbitrary attribute classifiers and satisfactory face-matching utility.
- 27 January 2025
- Published by Springer Nature in Multimedia Systems
- Vol. 31 (2) , 1-13
Segmentation-aware image super-resolution with generative adversarial networks
- 27 January 2025
- Published by Springer Nature in Multimedia Systems
- Vol. 31 (2) , 1-13
Segmentation-aware image super-resolution with generative adversarial networks
- Open Access
Block-Based Multi-Scale Image Rescaling
Image rescaling (IR) seeks to determine the optimal low-resolution (LR) representation of a high-resolution (HR) image to reconstruct a high-quality super-resolution (SR) image. Typically, HR images with resolutions exceeding 2K possess rich information that is unevenly distributed across the image. Traditional image rescaling methods often fall short because they focus solely on the overall scaling rate, ignoring the varying amounts of information in different parts of the image. To address this limitation, we propose a Block-Based Multi-Scale Image Rescaling Framework (BBMR), tailored for IR tasks involving HR images of 2K resolution and higher. BBMR consists of two main components: the Downscaling Module and the Upscaling Module. In the Downscaling Module, the HR image is segmented into sub-blocks of equal size, with each sub-block receiving a dynamically allocated scaling rate while maintaining a constant overall scaling rate. For the Upscaling Module, we introduce the Joint Super-Resolution method (JointSR), which performs SR on these sub-blocks with varying scaling rates and effectively eliminates blocking artifacts. Experimental results demonstrate that BBMR significantly enhances the SR image quality on the of 2K and 4K test dataset compared to initial network image rescaling methods.
Open AccessBlock-Based Multi-Scale Image Rescaling
Image rescaling (IR) seeks to determine the optimal low-resolution (LR) representation of a high-resolution (HR) image to reconstruct a high-quality super-resolution (SR) image. Typically, HR images with resolutions exceeding 2K possess rich information that is unevenly distributed across the image. Traditional image rescaling methods often fall short because they focus solely on the overall scaling rate, ignoring the varying amounts of information in different parts of the image. To address this limitation, we propose a Block-Based Multi-Scale Image Rescaling Framework (BBMR), tailored for IR tasks involving HR images of 2K resolution and higher. BBMR consists of two main components: the Downscaling Module and the Upscaling Module. In the Downscaling Module, the HR image is segmented into sub-blocks of equal size, with each sub-block receiving a dynamically allocated scaling rate while maintaining a constant overall scaling rate. For the Upscaling Module, we introduce the Joint Super-Resolution method (JointSR), which performs SR on these sub-blocks with varying scaling rates and effectively eliminates blocking artifacts. Experimental results demonstrate that BBMR significantly enhances the SR image quality on the of 2K and 4K test dataset compared to initial network image rescaling methods.
- Open Access
- 8 July 2024
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Sensor Networks
- Vol. 20 (4) , 1-24
Energy and QoE Optimization for Mobile Video Streaming with Adaptive Brightness Scaling
Brightness scaling (BS) is an emerging and promising technique with outstanding energy efficiency on mobile video streaming. However, existing BS-based approaches totally neglect the inherent interaction effect between BS factor, video bitrate and environment context. Their combined impact on user’s visual perception in mobile scenario, leading to inharmonious between energy consumption and user’s quality of experience (QoE). In this paper, we propose PEO , a novel user- P erception-based video E xperience O ptimization for energy-constrained mobile video streaming, by jointly considering the inherent connection between a device’s state of motion, video quality and the resulting user-perceived quality. Specifically, by capturing the motion of the on-the-run device, PEO first infers the optimal bitrate and BS factor, therefore avoiding bitrate-inefficiency for energy saving while guaranteeing the user-perceived QoE. On that basis, we formulate the device motion-aware and user perception-aware video streaming as an optimization problem where we present an optimal algorithm to maximize the object function and adapt to user preference, and thus propose an online bitrate selection algorithm. Our evaluation (based on trace analysis and user study) shows that, compared with state-of-the-art techniques, PEO can raise the perceived quality by 23.8%-41.3% and save up to 25.2% energy consumption.
- 8 July 2024
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Sensor Networks
- Vol. 20 (4) , 1-24
Open AccessEnergy and QoE Optimization for Mobile Video Streaming with Adaptive Brightness Scaling
Brightness scaling (BS) is an emerging and promising technique with outstanding energy efficiency on mobile video streaming. However, existing BS-based approaches totally neglect the inherent interaction effect between BS factor, video bitrate and environment context. Their combined impact on user’s visual perception in mobile scenario, leading to inharmonious between energy consumption and user’s quality of experience (QoE). In this paper, we propose PEO , a novel user- P erception-based video E xperience O ptimization for energy-constrained mobile video streaming, by jointly considering the inherent connection between a device’s state of motion, video quality and the resulting user-perceived quality. Specifically, by capturing the motion of the on-the-run device, PEO first infers the optimal bitrate and BS factor, therefore avoiding bitrate-inefficiency for energy saving while guaranteeing the user-perceived QoE. On that basis, we formulate the device motion-aware and user perception-aware video streaming as an optimization problem where we present an optimal algorithm to maximize the object function and adapt to user preference, and thus propose an online bitrate selection algorithm. Our evaluation (based on trace analysis and user study) shows that, compared with state-of-the-art techniques, PEO can raise the perceived quality by 23.8%-41.3% and save up to 25.2% energy consumption.
- 16 April 2024
- Published by Elsevier in Future Generation Computer Systems
- Vol. 158, 200-209
Region-based compressive distributed storage in Mobile CrowdSensing
- 16 April 2024
- Published by Elsevier in Future Generation Computer Systems
- Vol. 158, 200-209
Region-based compressive distributed storage in Mobile CrowdSensing
- 8 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 26 (15209210) , 7503-7515
Pedestrian Trajectory Prediction Based on Social Interactions Learning With Random Weights
Pedestrian trajectory prediction is a critical technology in the evolution of self-driving cars toward complete artificial intelligence. Over recent years, focusing on the trajectories of pedestrians to model their social interactions has surged with great interest in more accurate trajectory predictions. However, existing methods for modeling pedestrian social interactions rely on pre-defined rules, struggling to capture non-explicit social interactions. In this work, we propose a novel framework named DTGAN, which extends the application of Generative Adversarial Networks (GANs) to graph sequence data, with the primary objective of automatically capturing implicit social interactions and achieving precise predictions of pedestrian trajectory. DTGAN innovatively incorporates random weights within each graph to eliminate the need for pre-defined interaction rules. We further enhance the performance of DTGAN by exploring diverse task loss functions during adversarial training, which yields improvements of 16.7% and 39.3% on metrics ADE and FDE, respectively. The effectiveness and accuracy of our framework are verified on two public datasets. The experimental results show that our proposed DTGAN achieves superior performance and is well able to understand pedestrians' intentions.
- 8 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 26 (15209210) , 7503-7515
Pedestrian Trajectory Prediction Based on Social Interactions Learning With Random Weights
Pedestrian trajectory prediction is a critical technology in the evolution of self-driving cars toward complete artificial intelligence. Over recent years, focusing on the trajectories of pedestrians to model their social interactions has surged with great interest in more accurate trajectory predictions. However, existing methods for modeling pedestrian social interactions rely on pre-defined rules, struggling to capture non-explicit social interactions. In this work, we propose a novel framework named DTGAN, which extends the application of Generative Adversarial Networks (GANs) to graph sequence data, with the primary objective of automatically capturing implicit social interactions and achieving precise predictions of pedestrian trajectory. DTGAN innovatively incorporates random weights within each graph to eliminate the need for pre-defined interaction rules. We further enhance the performance of DTGAN by exploring diverse task loss functions during adversarial training, which yields improvements of 16.7% and 39.3% on metrics ADE and FDE, respectively. The effectiveness and accuracy of our framework are verified on two public datasets. The experimental results show that our proposed DTGAN achieves superior performance and is well able to understand pedestrians' intentions.