In this letter, we predict the locations as probability distributions for the tasks of image object detection. We adopt the Kullback-Leibler divergences as the regression losses to train the deep neural networks. Since most existing evaluations label the objects with rectangular bounding boxes, we propose the Nearest Distribution Converter to find the closest uniform distributions from the predicted ones. Our proposed method can improve the detected accuracy measured in mAP by 0.57%, 0.75%, and 0.48% on the models YOLOv3, the YOLOv4-tiny, and the YOLOv4, respectively.

Two-way full-duplex (TWFD) transmissions in cooperative amplify-and-forward (AF) relaying systems provide a promising way of communications with the improved spectral efficiency by concurrently transmitting/receiving data in the same frequency band and by the two-way transmission protocol. However, the loopback self-interference, caused by the imperfect channel estimation in the loopback link, at the AF relay accumulates and is forwarded to the end users recurrently, which substantially degrades the end users' error performance if this accumulated loopback self-interference (ALSI) does not converge. It is therefore of critical importance to analyze the convergence behavior of the ALSI in the TWFD systems with AF relaying. In this paper, we establish sufficient conditions on the relay power to ensure the convergence of the ALSI, with which the average limiting signal-to-interference-plus-noise ratio at the infinite time horizon is derived. In addition, we specify the relay power at which the signal-to-noise ratio reaches the maximal value under different scenarios and obtain an upper bound to the symbol error rate. Finally, extensive computer simulations results are provided to corroborate the analytic derivations.

Rate control plays an important role in video coding and streaming applications with bandwidth constraints. While most researches are proposed to improve the coding efficiency' the fluctuation of video quality is seldom considered. Many rate control schemes suffer from unreliable initialization of coding parameters, which leads to seriously inconsistent quality at the beginning of a video. Besides, the hierarchical structure for frame references introduces more quality fluctuations, although it improves the coding efficiency significantly. This paper presents a two pass rate control method that aims for a consistent visual quality. The video is down-sampled by four times, and then encoded for the first pass. A fixed Lagrange multiplier (λ) is derived from the information recorded in the first pass, and then applied for all frames in the second coding pass. A QP adjustment policy is adopted to maintain a consistent quality and a constant bitrate. Experimental results show that the proposed rate control method can reduce the fluctuation of video quality to be averagely 94.63% less than that encoded by the HEVC Test Model (HM16.9).

In this paper, we propose to improve the performance of the rate-control mechanism in video encoders by adaptively adjusting the Lagrange multiplier λ based on the stochastic subgradient method. Using this λ value for each coding unit, the quantization parameter (QP) is sought in the process of rate-distortion optimization (RDO). A log-linear model can be utilized to predict the most probable QP and accordingly narrow the search range. The generated bit number is used to update the λ value for the next coding unit. The extensive experimental results show that the proposed rate-control mechanism significantly enhances the rate-distortion performance compared with the H.264/AVC (Advanced Video Coding) reference software.

In this paper, we propose a high-precision digital framework to compensate for non-ideal effects in the analog front-end circuits for wireless communication systems. The IQ imbalance and the carrier frequency offset (CFO) are jointly estimated with a band-selective mechanism of the orthogonal frequency division multiplexing (OFDM) system. The compensable CFO range is four times wider than prior schemes. The proposed mechanism divides the whole range of CFO bandwidth into three bands: the high-CFO band, the medium-CFO band, and the low-CFO band. The least-square (LS) based method is performed to simultaneously estimate the IQ imbalance and the CFO. Simulation results show that the proposed band-selective algorithm provides accurate estimation, with the average error less than 0.13%.

A low-complexity divider-free architecture is pro- posed to implement the direct prediction technique of sub- pixel motion estimation. We investigate the characteristic of the error surface to reduce the computation complexity. The flatness of the error surface around the best integer-pixel motion vector (MV) position is measured to determine whether the prediction is skipped to save computation. The dividers are replaced by comparators to shorten the critical path. We also develop a simplified process to find the best prediction inside the well-condition error surface. Experimental results show that the performance of the proposed low-complexity sub-pixel MV prediction scheme is close to the conventional full interpolation- and -search method.

This paper jointly considers the mechanisms of rate control and the rate-distortion optimization (RDO) for the H.264/AVC hardware encoder. The proposed architecture performs intra prediction and inter prediction with our rate control mechanism in the first two stages. In addition, a low complexity macroblock (MB)-level mean absolute difference (MAD) prediction algorithm which considers both the motion information and the MAD value in the neighboring and current MBs is proposed. The proposed rate control algorithm can reduce the computational complexity and does not suffer from data dependency problems which may decrease the rate-distortion performance, thus it is suitable for H.264/AVC hardware encoders. Furthermore, the RDO-based architecture for the H.264/AVC encoder reduces the number of accurate rate-distortion cost calculators while maintaining the high rate-distortion performance. Experimental results show that the proposed rate control algorithm with the RDO-based architecture can improve the rate-distortion performance of H.264/AVC.

A special type of cooperative wireless transmission scheme is proposed for the broadcasting of scalable video sources. In the proposed system, a transmitter with multiple antennas sends out signals by encoding layered space-time codes. The receiver alone without cooperation can decode the base layer to get the basic quality of the video. For the cooperative terminals, with the help of relayed information, the enhanced layers can be retrieved and then the visual quality can be refined. Simulation results with video bitstreams encoded by H.264/scalable video coding show that the proposed system can enhance the scalable functionality of the video.

This paper presents two low complexity macroblock-level MAD (Mean Absolute Difference) prediction algorithms for H.264/AVC rate control. Both of the algorithms consider the rate-distortion performance and its hardware cost jointly. One of the algorithms predicts the MAD value by observing the motion vector of the surrounding macroblocks (MB), the other one uses numerical analysis to find the best fit weighting factors for the MAD value. These algorithms do not suffer from the MB-pipeline problems which decrease the rate-distortion performance, thus they are suitable for H.264/AVC hardware encoders. Experimental results show that the proposed MAD prediction algorithm can further improve the rate-distortion performance for the rate control scheme. Compared with the JVT-G012 rate control scheme, our algorithms can increase the PSNR to 0.38 dB and 0.35 dB, respectively.

We investigate a commonly used power control scheme called single-step power control (SSPC) in finite state Markov channel (FSMC). The performance of SSPC is highly affected by two major factors, the power control error (PCE) and the imperfect feedback of the SSPC commands. In the first part of this work, we aim at modelling the behavior of PCE in FSMC. Based on the proposed PCE model, BER and PER performance are evaluated. Simulation results have shown the validity of the proposed PCE model, derived PER and BER in this work. In the second part of this work, SSPC models under different scenarios are established in order to understand the impact of the imperfect feedback on SSPC commands and to obtain the interactions between SSPC and FSMC. We consider four scenarios, perfect feedback, command error, command loss and command delay. The associated transitional probabilities of each scenario are given. Extensive simulation has shown the accuracy of the proposed SSPC models in these four scenarios.

This paper presents a low-complexity rate control (RC) framework for H.264/advanced video coding (AVC) encoders on embedded systems. We consider not only the rate-distortion performance but also its implementation complexity using an efficient hardware/software codesign. The proposed cost-effective macroblock level RC module works in hardware, and the frame-level RC is executed on the system CPU as software. It is less complex than the RC module adopted by H.264/AVC JM reference software, making it more suitable for embedded systems. Experiment results show that the proposed RC outperforms H.264/AVC JM, and the hardware implementation cost is low for an H.264/AVC video encoder.

To improve video quality in TV display, we propose an architecture including frame rate up-conversion (FRUC), spatial scaling, and adaptively edge sharpening. The technique of motion compensation is applied both to frame interpolation and to edge sharpening in order to improve visual quality. In conventional methods, the bi-directional motion estimation (ME) is usually employed between two successive frames to interpolate a new frame. The proposed method not only considers temporal correlation, but also exploits the spatial correlation to determine the covered and uncovered region. The weighting of edge sharpness is determined by the magnitude of motion vector (MV) and mean absolute difference (MAD) of the current block to improve image quality. We implement and verify our architecture on a virtual platform with system level design flow. Simulation results show that our method has better PSNR and image quality.

In this paper, we propose a real time hardware architecture with low cost for histogram difference calculator to process and low-level extraction in our video abstraction system. The highlights are extracted from video data by directly combining simple visual and audio features of video data without specific domain knowledge and hence its low complexity makes it suitable to be implemented in embedded systems. Experimental results show that the condensed skimming clips comprise the interesting and informative parts and most meaningful semantic contents are extracted efficiently

A fast inter-prediction mode decision and motion search algorithm is proposed for the H.264 video coding standard. The multi-resolution motion estimation scheme and an adaptive rate-distortion model are employed with early termination rules to accelerate the search process. With the new algorithm, the amount of computation involved in the motion search can be substantially reduced. Experimental results show that the proposed algorithm can achieve a speed-up factor ranging from 60 to 150 times as compared to the full-search algorithm with little quality degradation.

An enhanced version of the fast inter-prediction mode decision and variable block size motion estimation scheme for the H.264 video encoder is proposed in this work. In particular, we focus on the rate estimation model and thresholds for early termination. They are adaptively updated to enhance the coding speed and the rate-distortion performance. More details about the threshold setting and the mode decision processes are also provided. The improved algorithm can achieve a speed-up factor ranging from 70 to 150 times as compared to the full-search algorithm.

An embedded space-time coding method is proposed for wireless broadcast with heterogeneous receivers. In the proposed system, a transmitter sends out multi-layer source signals by encoding different layers with different space-time codes. Then, a receiver can retrieve a different amount of information depending on the number of its antennas. The receiver with only one antenna can decode only the base layer information with a low complexity, while the receiver with more antennas can retrieve more layers of information. We also investigate a differential design of the embedded space-time system, which is used when the channel state information is unknown to the receiver. Both analytic and experimental results are provided to demonstrate the performance of the proposed system.

A space-time coded orthogonal frequency-division multiplexed (STC-OFDM) system is proposed to transmit layered video signals over wireless channels. The space-time coding can provide a diversity gain for the multiple-antenna system while OFDM can alleviate the frequency-selective fading effect. An input video sequence is compressed and partitioned into layers with different priorities. Then, an unequal error protection scheme based on the product Reed-Solomon (RS) codes is employed to provide different levels of protection to different layers. At the receiver, a minimum mean square error (MMSE) detector with interference cancellation (IC) is integrated with the space-time decoder to reconstruct the desired signal effectively. We derive an analytic bound for the error probability, and conduct experiments for the transmission of the H.263 video bitstream. It is shown that the proposed scheme enhances the PSNR performance significantly in the environment with a moderate signal-to-noise ratio.