Zhifeng Chen
Research Subjects
Zhifeng Chen has a total of 40 co-authors. Here are the top co-authors by number of shared publications:
- Dapeng Oliver Wu
City University of Hong Kong
5 shared publications - Pingping Chen
Fuzhou University
3 shared publications - Bo Huang
Beijing Kuaishou Technology Ltd., Beijing, 100092 China
2 shared publications - Ente Guo
School of Computer and Big Data, Minjiang University, Fuzhou 350108, China
2 shared publications - Mingkui Zheng
Fuzhou University
2 shared publications - Qi Cai
Department of Electrical and Computer Engineering, University of Florida, Gainesville, FL, USA
2 shared publications
Latest Publications
- 1 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 34 (5) , 4054-4069
Camera Pose-Based Background Modeling for Video Coding in Moving Cameras
For moving cameras, the video content changes significantly, which leads to inaccurate prediction in traditional inter prediction and results in limited compression efficiency. To solve these problems, first, we propose a camera pose-based background modeling (CP-BM) framework that uses the camera motion and the textures of reconstructed frames to model the background of the current frame. Compared with the reconstructed frames, the predicted background frame generated by CP-BM is more geometrically similar to the current frame in position and is more strongly correlated with it at the pixel level; thus, it can serve as a higher-quality reference for inter prediction, and the compression efficiency can be improved. Second, to compensate the motion of the background pixels, we construct a pixel-level motion vector field that can accurately describe various complex motions with only a small overhead. Our method is more general than other motion models because it has more degrees of freedom, and when the degrees of freedom are decreased, it encompasses other motion models as special cases. Third, we propose an optical flow-based depth estimation (OF-DE) method to synchronize the depth information at the codec, which is used to build the motion vector field. Finally, we integrate the overall scheme into the High Efficiency Video Coding (HEVC) and Versatile Video Coding (VVC) reference software HM-16.7 and VTM-10.0. Experimental results demonstrate that in HM-16.7, for in-vehicle video sequences, our solution has an average Bjøntegaard delta bit rate (BD-rate) gain of 8.02% and reduces the encoding time by 20.9% due to the superiority of our scheme in motion estimation. Moreover, in VTM-10.0 with affine motion compensation (MC) turned off and turned on, our method has average BD-rate gains of 5.68% and 0.56%, respectively.
- 1 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 34 (5) , 4054-4069
Camera Pose-Based Background Modeling for Video Coding in Moving Cameras
For moving cameras, the video content changes significantly, which leads to inaccurate prediction in traditional inter prediction and results in limited compression efficiency. To solve these problems, first, we propose a camera pose-based background modeling (CP-BM) framework that uses the camera motion and the textures of reconstructed frames to model the background of the current frame. Compared with the reconstructed frames, the predicted background frame generated by CP-BM is more geometrically similar to the current frame in position and is more strongly correlated with it at the pixel level; thus, it can serve as a higher-quality reference for inter prediction, and the compression efficiency can be improved. Second, to compensate the motion of the background pixels, we construct a pixel-level motion vector field that can accurately describe various complex motions with only a small overhead. Our method is more general than other motion models because it has more degrees of freedom, and when the degrees of freedom are decreased, it encompasses other motion models as special cases. Third, we propose an optical flow-based depth estimation (OF-DE) method to synchronize the depth information at the codec, which is used to build the motion vector field. Finally, we integrate the overall scheme into the High Efficiency Video Coding (HEVC) and Versatile Video Coding (VVC) reference software HM-16.7 and VTM-10.0. Experimental results demonstrate that in HM-16.7, for in-vehicle video sequences, our solution has an average Bjøntegaard delta bit rate (BD-rate) gain of 8.02% and reduces the encoding time by 20.9% due to the superiority of our scheme in motion estimation. Moreover, in VTM-10.0 with affine motion compensation (MC) turned off and turned on, our method has average BD-rate gains of 5.68% and 0.56%, respectively.
- 14 September 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-4
Artificial Neural Networks Modeling Technology for Substrate Integrated Suspended Line
A new method for parameter modeling of self-packaged Substrate Integrated Suspended Line (SISL) devices based on artificial neural network is proposed. The method only uses the input and output signals of self-packaged SISL devices, and does not need to establish the internal structure information of the model. The cost of modeling is greatly reduced and the design cycle of microwave passive devices is shortened. In addition, an improved training method is proposed based on the neural network to learn the relationship between electromagnetic response and geometric parameters, so that the proposed model can efficiently and accurately match the characteristics of SISL devices. The model takes geometric parameters as variables. It can provide the exact and rapid prediction for the electromagnetic response of SISL devices. The validity and accuracy of the parameter modeling technique based on an artificial neural network are verified by a modeling example of a wideband Yagi antenna.
- 14 September 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-4
Artificial Neural Networks Modeling Technology for Substrate Integrated Suspended Line
A new method for parameter modeling of self-packaged Substrate Integrated Suspended Line (SISL) devices based on artificial neural network is proposed. The method only uses the input and output signals of self-packaged SISL devices, and does not need to establish the internal structure information of the model. The cost of modeling is greatly reduced and the design cycle of microwave passive devices is shortened. In addition, an improved training method is proposed based on the neural network to learn the relationship between electromagnetic response and geometric parameters, so that the proposed model can efficiently and accurately match the characteristics of SISL devices. The model takes geometric parameters as variables. It can provide the exact and rapid prediction for the electromagnetic response of SISL devices. The validity and accuracy of the parameter modeling technique based on an artificial neural network are verified by a modeling example of a wideband Yagi antenna.
- 30 July 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 21 (19) , 21759-21771
3D Detection and Pose Estimation of Vehicle in Cooperative Vehicle Infrastructure System
Three-dimensional (3D) object detection is of great significance for avoiding collisions between vehicles and obstacles in autonomous driving. In particular, the recent 3D object detection methods based on supervised learning are widely studied to achieve excellent performance. However, the 3D labels for training in such methods are expensive and often difficult to be collected. To solve this issue, we propose a monocular 3D vehicle detection method. First, we propose a general mathematical K-means-like method for clustering arbitrary object contours into linear equations. Second, the position, orientation and dimensions of the vehicle can be estimated by applying K-means-like method without the need for 3D labels in the contour of the vehicle. Finally, given the 2D object detection, we maximize a posterior probability of vehicle position, orientation and dimensions to improve the accuracy of the 3D object detection based on the results of K-means-like method. We evaluate the proposed algorithm on the dataset collected by the vehicle-side and road-side cameras in the cooperative vehicle infrastructure system (CVIS). Compared with the state-of-art Deep3DBox and SMOKE methods, the evaluated results show that the detection accuracy of 3D object of our method is 1.4% higher than that of Deep3DBox in the vehicle-side system, while for the road-side camera, the proposed method has 3.86% and 4.37% higher accuracy than Deep3DBox and SMOKE, respectively. Thus, the proposed method can be seen as an effective 3D object detection method in the intelligent transportation system and CVIS.
- 30 July 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 21 (19) , 21759-21771
3D Detection and Pose Estimation of Vehicle in Cooperative Vehicle Infrastructure System
Three-dimensional (3D) object detection is of great significance for avoiding collisions between vehicles and obstacles in autonomous driving. In particular, the recent 3D object detection methods based on supervised learning are widely studied to achieve excellent performance. However, the 3D labels for training in such methods are expensive and often difficult to be collected. To solve this issue, we propose a monocular 3D vehicle detection method. First, we propose a general mathematical K-means-like method for clustering arbitrary object contours into linear equations. Second, the position, orientation and dimensions of the vehicle can be estimated by applying K-means-like method without the need for 3D labels in the contour of the vehicle. Finally, given the 2D object detection, we maximize a posterior probability of vehicle position, orientation and dimensions to improve the accuracy of the 3D object detection based on the results of K-means-like method. We evaluate the proposed algorithm on the dataset collected by the vehicle-side and road-side cameras in the cooperative vehicle infrastructure system (CVIS). Compared with the state-of-art Deep3DBox and SMOKE methods, the evaluated results show that the detection accuracy of 3D object of our method is 1.4% higher than that of Deep3DBox in the vehicle-side system, while for the road-side camera, the proposed method has 3.86% and 4.37% higher accuracy than Deep3DBox and SMOKE, respectively. Thus, the proposed method can be seen as an effective 3D object detection method in the intelligent transportation system and CVIS.
- 18 May 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Broadcasting
- Vol. 67 (3) , 721-735
Low-Complexity Rate-Distortion Optimization for HEVC Encoders
Typical high efficiency video coding (HEVC) encoders use rate-distortion optimization (RDO) to select the best coding parameters among numerous candidates to achieve high compression efficiency. However, calculating the rate-distortion cost for each coding parameter entails high computational complexity, which limits the real-time application of these encoders. In this study, time-consuming processes, such as reconstruction and entropy coding, are removed during the calculation of the full RD cost to reduce the complexity of the encoder. Instead, RD costs are estimated for transformed coefficients by using the RD cost model, which is made up of a distortion model based on the distribution features of the transformed coefficients, a residual rate model based on the characteristics of the entropy coding module, and a simple and convenient header rate model. In addition to the full RD cost, typical encoders also adopt fast RD cost, which is composed of the sum of absolute transform differences (SATD) and header bit. To further reduce encoding complexity, three simplified SATD (SSATD) methods based on DCT and DST features are proposed to serve as substitutes for SATD in fast RD cost, thus reducing the computational complexity of SATD calculation. Experimental results show that 35.53%, 14.57%, and 14.37% of coding time can be reduced respectively under AI, RA and LDP configurations with negligible RD performance loss when applying the proposed RD cost model and the SSATD methods to HM16.7. Besides, the proposed methods also can reduce 4%-9% of encoding time and exhibits the same RD performance as state-of-the-art methods.
- 18 May 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Broadcasting
- Vol. 67 (3) , 721-735
Low-Complexity Rate-Distortion Optimization for HEVC Encoders
Typical high efficiency video coding (HEVC) encoders use rate-distortion optimization (RDO) to select the best coding parameters among numerous candidates to achieve high compression efficiency. However, calculating the rate-distortion cost for each coding parameter entails high computational complexity, which limits the real-time application of these encoders. In this study, time-consuming processes, such as reconstruction and entropy coding, are removed during the calculation of the full RD cost to reduce the complexity of the encoder. Instead, RD costs are estimated for transformed coefficients by using the RD cost model, which is made up of a distortion model based on the distribution features of the transformed coefficients, a residual rate model based on the characteristics of the entropy coding module, and a simple and convenient header rate model. In addition to the full RD cost, typical encoders also adopt fast RD cost, which is composed of the sum of absolute transform differences (SATD) and header bit. To further reduce encoding complexity, three simplified SATD (SSATD) methods based on DCT and DST features are proposed to serve as substitutes for SATD in fast RD cost, thus reducing the computational complexity of SATD calculation. Experimental results show that 35.53%, 14.57%, and 14.37% of coding time can be reduced respectively under AI, RA and LDP configurations with negligible RD performance loss when applying the proposed RD cost model and the SSATD methods to HM16.7. Besides, the proposed methods also can reduce 4%-9% of encoding time and exhibits the same RD performance as state-of-the-art methods.
- 1 February 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 31 (12) , 4924-4937
A Novel Video Coding Strategy in HEVC for Object Detection
Occupying the most significant portion of global data traffic, video is being generated in almost every aspect of our life. Because of its huge volume, we are depending much more heavily on machine intelligence based analysis. In the meantime, video coding technology has been continuously improved for better compression efficiency. However, the state-of-the-art video coding standards, such as H.265/HEVC and versatile video coding (VVC), are still designed assuming that the compressed video will be watched by a human later. Such a design is not optimal when the compressed video will be used by computer vision applications. While the human visual system (HVS) is consistently sensitive to the content with high contrast, the impact of pixels on computer vision algorithms is task driven. For example, because of the different categories of objects used to train detection algorithms, the influence of the same image content on those detectors also varies. Therefore, human oriented video coding strategies may not be optimal when the compressed signal is further processed by algorithms, as the encoder is unaware of the task specific information. In this article, taking object detection as an example, we propose a novel video coding strategy for computer vision. By protecting the information according to its importance for an object detector rather than for the human visual system, our proposed method has the potential to achieve a better object detection performance with the same bandwidth. The main contributions of our paper are: 1) the modeling of the relationship between object detection accuracy and bit rate; 2) a back propagation based method to analyze the influence of each pixel on the detection of target objects; 3) an object detection oriented bit allocation and codec control parameter determination scheme; 4) an evaluation metric to compare the impact of video coding strategies on a given object detector over a predefined range of bit rate. Experimental results demonstrate that our proposed algorithm can better preserve the video content vital for object detection than state-of-the-art video coding schemes.
- 1 February 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 31 (12) , 4924-4937
A Novel Video Coding Strategy in HEVC for Object Detection
Occupying the most significant portion of global data traffic, video is being generated in almost every aspect of our life. Because of its huge volume, we are depending much more heavily on machine intelligence based analysis. In the meantime, video coding technology has been continuously improved for better compression efficiency. However, the state-of-the-art video coding standards, such as H.265/HEVC and versatile video coding (VVC), are still designed assuming that the compressed video will be watched by a human later. Such a design is not optimal when the compressed video will be used by computer vision applications. While the human visual system (HVS) is consistently sensitive to the content with high contrast, the impact of pixels on computer vision algorithms is task driven. For example, because of the different categories of objects used to train detection algorithms, the influence of the same image content on those detectors also varies. Therefore, human oriented video coding strategies may not be optimal when the compressed signal is further processed by algorithms, as the encoder is unaware of the task specific information. In this article, taking object detection as an example, we propose a novel video coding strategy for computer vision. By protecting the information according to its importance for an object detector rather than for the human visual system, our proposed method has the potential to achieve a better object detection performance with the same bandwidth. The main contributions of our paper are: 1) the modeling of the relationship between object detection accuracy and bit rate; 2) a back propagation based method to analyze the influence of each pixel on the detection of target objects; 3) an object detection oriented bit allocation and codec control parameter determination scheme; 4) an evaluation metric to compare the impact of video coding strategies on a given object detector over a predefined range of bit rate. Experimental results demonstrate that our proposed algorithm can better preserve the video content vital for object detection than state-of-the-art video coding schemes.
- Open Access
Unsupervised Learning of Depth and Camera Pose with Feature Map Warping
Estimating the depth of image and egomotion of agent are important for autonomous and robot in understanding the surrounding environment and avoiding collision. Most existing unsupervised methods estimate depth and camera egomotion by minimizing photometric error between adjacent frames. However, the photometric consistency sometimes does not meet the real situation, such as brightness change, moving objects and occlusion. To reduce the influence of brightness change, we propose a feature pyramid matching loss (FPML) which captures the trainable feature error between a current and the adjacent frames and therefore it is more robust than photometric error. In addition, we propose the occlusion-aware mask (OAM) network which can indicate occlusion according to change of masks to improve estimation accuracy of depth and camera pose. The experimental results verify that the proposed unsupervised approach is highly competitive against the state-of-the-art methods, both qualitatively and quantitatively. Specifically, our method reduces absolute relative error (Abs Rel) by 0.017–0.088.
Open AccessUnsupervised Learning of Depth and Camera Pose with Feature Map Warping
Estimating the depth of image and egomotion of agent are important for autonomous and robot in understanding the surrounding environment and avoiding collision. Most existing unsupervised methods estimate depth and camera egomotion by minimizing photometric error between adjacent frames. However, the photometric consistency sometimes does not meet the real situation, such as brightness change, moving objects and occlusion. To reduce the influence of brightness change, we propose a feature pyramid matching loss (FPML) which captures the trainable feature error between a current and the adjacent frames and therefore it is more robust than photometric error. In addition, we propose the occlusion-aware mask (OAM) network which can indicate occlusion according to change of masks to improve estimation accuracy of depth and camera pose. The experimental results verify that the proposed unsupervised approach is highly competitive against the state-of-the-art methods, both qualitatively and quantitatively. Specifically, our method reduces absolute relative error (Abs Rel) by 0.017–0.088.
- Open Access
- 7 April 2020
- Published by MDPI AG in Applied Sciences
- Vol. 10 (7) , 2533
Infrared Image Deblurring via High-Order Total Variation and Lp-Pseudonorm Shrinkage
The quality of infrared images is affected by various degradation factors, such as image blurring and noise pollution. Anisotropic total variation (ATV) has been shown to be a good regularization approach for image deblurring. However, there are two main drawbacks in ATV. First, the conventional ATV regularization just considers the sparsity of the first-order image gradients, thus leading to staircase artifacts. Second, it employs the L1-norm to describe the sparsity of image gradients, while the L1-norm has a limited capacity of depicting the sparsity of sparse variables. To address these limitations of ATV, a high-order total variation is introduced in the ATV deblurring model and the Lp-pseudonorm is adopted to depict the sparsity of low- and high-order total variation. In this way, the recovered image can fit the image priors with clear edges and eliminate the staircase artifacts of the ATV model. The alternating direction method of multipliers is used to solve the proposed model. The experimental results demonstrate that the proposed method does not only remove blurs effectively but is also highly competitive against the state-of-the-art methods, both qualitatively and quantitatively.
- 7 April 2020
- Published by MDPI AG in Applied Sciences
- Vol. 10 (7) , 2533
Open AccessInfrared Image Deblurring via High-Order Total Variation and Lp-Pseudonorm Shrinkage
The quality of infrared images is affected by various degradation factors, such as image blurring and noise pollution. Anisotropic total variation (ATV) has been shown to be a good regularization approach for image deblurring. However, there are two main drawbacks in ATV. First, the conventional ATV regularization just considers the sparsity of the first-order image gradients, thus leading to staircase artifacts. Second, it employs the L1-norm to describe the sparsity of image gradients, while the L1-norm has a limited capacity of depicting the sparsity of sparse variables. To address these limitations of ATV, a high-order total variation is introduced in the ATV deblurring model and the Lp-pseudonorm is adopted to depict the sparsity of low- and high-order total variation. In this way, the recovered image can fit the image priors with clear edges and eliminate the staircase artifacts of the ATV model. The alternating direction method of multipliers is used to solve the proposed model. The experimental results demonstrate that the proposed method does not only remove blurs effectively but is also highly competitive against the state-of-the-art methods, both qualitatively and quantitatively.
- Open Access
- 6 January 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 52759-52773
On Consideration of Content and Memory Sizes in 5G D2D-Assisted Caching Networks
Device-to-Device (D2D) communication with caching technology has emerged as a promising technique for offloading the traffic and boosting the throughput of the fifth generation (5G) cellular networks. The combined impact of cache memory size of user equipments (UEs) and content sizes, which are two crucial factors in D2D-assisted caching networks, were usually ignored in the existing researches. In this paper, an optimization algorithm is proposed to maximize the cache hit probability and cache-aided throughput, with the consideration of various cache memory sizes of UEs and content sizes. Firstly, users are grouped according to the cache memory sizes and the content sizes. Then the general mathematical expressions for the optimization of cache hit probability and cache-aided throughput with the constraints of cache memory sizes and content sizes are obtained. Subsequently, a Packet Cache Strategy (PCS) algorithm is proposed to obtain the caching probability matrix with the maximum cache hit probability and cache-aided throughout by taking user caching probability of a file as a variable. Finally, numerical results show that the sizes of the requested files affect the caching willingness of users, and the proposed PCS can achieve the highest cache hit probability and the best cache-aided throughput comparing with two other existing methods.
- 6 January 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 52759-52773
Open AccessOn Consideration of Content and Memory Sizes in 5G D2D-Assisted Caching Networks
Device-to-Device (D2D) communication with caching technology has emerged as a promising technique for offloading the traffic and boosting the throughput of the fifth generation (5G) cellular networks. The combined impact of cache memory size of user equipments (UEs) and content sizes, which are two crucial factors in D2D-assisted caching networks, were usually ignored in the existing researches. In this paper, an optimization algorithm is proposed to maximize the cache hit probability and cache-aided throughput, with the consideration of various cache memory sizes of UEs and content sizes. Firstly, users are grouped according to the cache memory sizes and the content sizes. Then the general mathematical expressions for the optimization of cache hit probability and cache-aided throughput with the constraints of cache memory sizes and content sizes are obtained. Subsequently, a Packet Cache Strategy (PCS) algorithm is proposed to obtain the caching probability matrix with the maximum cache hit probability and cache-aided throughout by taking user caching probability of a file as a variable. Finally, numerical results show that the sizes of the requested files affect the caching willingness of users, and the proposed PCS can achieve the highest cache hit probability and the best cache-aided throughput comparing with two other existing methods.
- Open Access
Physical-Layer Network Coding: An Efficient Technique for Wireless Communications
As a subfield of network coding, physical-layer network coding (PNC) can effectively enhance the throughput of wireless networks by mapping superimposed signals at receiver to other forms of user messages. Over the past twenty years, PNC has received significant research attention and has been widely studied in various communication scenarios, e.g., two-way relay communications (TWRC), nonorthogonal multiple access (NOMA) in 5G networks, random access networks, etc. To ensure network reliability, channel-coded PNC is proposed and related communication techniques are investigated, such as the design of channel code, low-complexity decoding, and cross-layer design. In this article, we briefly review the variants of channel-coded PNC wireless communications with the aim of inspiring future research activities in this area. We also put forth open research problems along with a few selected research directions under PNC-aided frameworks.
Open AccessPhysical-Layer Network Coding: An Efficient Technique for Wireless Communications
As a subfield of network coding, physical-layer network coding (PNC) can effectively enhance the throughput of wireless networks by mapping superimposed signals at receiver to other forms of user messages. Over the past twenty years, PNC has received significant research attention and has been widely studied in various communication scenarios, e.g., two-way relay communications (TWRC), nonorthogonal multiple access (NOMA) in 5G networks, random access networks, etc. To ensure network reliability, channel-coded PNC is proposed and related communication techniques are investigated, such as the design of channel code, low-complexity decoding, and cross-layer design. In this article, we briefly review the variants of channel-coded PNC wireless communications with the aim of inspiring future research activities in this area. We also put forth open research problems along with a few selected research directions under PNC-aided frameworks.
- Open Access
- 26 February 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 7, 27401-27415
Buffer-Driven Rate Control and Packet Distribution for Real-Time Videos in Heterogeneous Wireless Networks
- 26 February 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 7, 27401-27415
Open AccessBuffer-Driven Rate Control and Packet Distribution for Real-Time Videos in Heterogeneous Wireless Networks