探索低资源的迭代式复述生成增强方法

复述生成旨在同一语言内将给定句子转换成语义一致表达不同的句子。目前,基于深度神经网络的复述生成模型的成功依赖于大规模的复述平行语料,当面向新的语言或新的领域时,模型性能急剧下降。面对这一困境,提出低资源的迭代式复述生成增强方法,最大化利用单语语料和小规模复述平行语料迭代式训练复述生成模型并生成复述伪数据,以此增强模型性能。此外,提出了句子流畅性、语义相近性和表达多样性为基准设计的伪数据筛选算法,选取高质量的复述伪数据参与每轮模型的迭代训练。在公开数据集Quora上的实验结果表明,提出的方法仅利用30%的复述语料在语义和多样性指标上均超过了基线模型,验证了所提方法的有效性。

基于碎片模型的弱纹理物体位姿估计

位姿估计已广泛应用于智能机器人,无人驾驶以及增强现实等多种应用场景,现有基于神经网路的点对匹配方法大多未能处理好前景遮挡,弱纹理以及对称。本文提出了一种基于碎片模型的弱纹理物体位姿估计方法,该方法利用现有的3D碎片模型从一幅RGB输入图像中估计出刚体物体的6D姿态;物体由紧凑表面碎片表示,能够系统地处理物体对称;使用编码器–解码器网络预测密集采样的像素和片段之间的对应关系;还提出了一种通用的数据合成方案,创建了高相似度的弱纹理物体数据集;最后,设计并改进了一种PNP-RANSAC算法稳健而有效地估计可能多个对象实例的姿态,并在3个弱纹理数据集上进行对比实验并验证了该方法的有效性。

基于关键词指导的图像中文描述生成

图像描述生成技术可以加速图文内容的生产,因而有着广泛的应用前景。为了满足实际需要,我们提出了一种基于编码–解码框架的新方法。我们的模型通过融合图像和文本特征作为输入来指导图像描述的生成,文本特征包含图像的关键词信息作为图像特征的补充。实验结果表明,关键词信息加强了图像到图像描述的映射,本文模型比未融合关键词信息的模型具有更好的性能,并且不同的关键词信息对图像描述的生成有一定的控制作用。

General and robust voxel feature learning with Transformer for 3D object detection

The self-attention networks and Transformer have dominated machine translation and natural language processing fields,and shown great potential in image vision tasks such as image classification and object detection.Inspired by the great progress of Transformer,we propose a novel general and robust voxel feature encoder for 3D object detection based on the traditional Transformer.We first investigate the permutation invariance of sequence data of the self-attention and apply it to point cloud processing.Then we construct a voxel feature layer based on the self-attention to adaptively learn local and robust context of a voxel according to the spatial relationship and context information exchanging between all points within the voxel.Lastly,we construct a general voxel feature learning framework with the voxel feature layer as the core for 3D object detection.The voxel feature with Transformer(VFT)can be plugged into any other voxel-based 3D object detection framework easily,and serves as the backbone for voxel feature extractor.Experiments results on the KITTI dataset demonstrate that our method achieves the state-of-the-art performance on 3D object detection.

CHANNEL ESTIMATION FOR ITERATIVE DECODING OVER FADING CHANNELS

A method of coherent detection and channel estimation for punctured convolutional coded binary Quadrature Amplitude Modulation (QAM) signals transmitted over a frequency-flat Rayleigh fading channels used for a digital radio broadcasting transmission is presented. Some known symbols are inserted in the encoded data stream to enhance the channel estimation process.The pilot symbols are used to replace the existing parity symbols so no bandwidth expansion is required. An iterative algorithm that uses decoding information as well as the information contained in the known symbols is used to improve the channel parameter estimate. The scheme complexity grows exponentially with the channel estimation filter length. The performance of the system is compared for a normalized fading rate with both perfect coherent detection (corresponding to a perfect knowledge of the fading process and noise variance) and differential detection of Differential Amplitude Phase Shift Keying (DAPSK). The tradeoff between simplicity of implementation and bit-error-rate performance of different techniques is also compared.

Frame-bitrate-change based steganography for voice-over-IP

Steganography based on bits-modification of speech frames is a kind of commonly used method, which targets at RTP payloads and offers covert communications over voice-over-IP(Vo IP). However, direct modification on frames is often independent of the inherent speech features, which may lead to great degradation of speech quality. A novel frame-bitrate-change based steganography is proposed in this work, which discovers a novel covert channel for Vo IP and introduces less distortion. This method exploits the feature of multi-rate speech codecs that the practical bitrate of speech frame is identified only by speech decoder at receiving end. Based on this characteristic, two steganography strategies called bitrate downgrading(BD) and bitrate switching(BS)are provided. The first strategy substitutes high bit-rate speech frames with lower ones to embed secret message, which introduces very low distortion in practice, and much less than other bits-modification based methods with the same embedding capacity. The second one encodes secret message bits into different types of speech frames, which is an alternative choice for supplement. The two strategies are implemented and tested on our covert communication system Steg Vo IP. The experiment results show that our proposed method is effective and fulfills the real-time requirement of Vo IP communication.

Optical encryption with cascaded fractional wavelet transforms

On the basis of fractional wavelet transform, we propose a new method called cascaded fractional wavelet transform to encrypt images. It has the virtues of fractional Fourier transform and wavelet transform. Fractional orders, standard focal lengths and scaling factors are its keys. Multistage fractional Fourier transforms can add the keys easily and strengthen information se-curity. This method can also realize partial encryption just as wavelet transform and fractional wavelet transform. Optical reali-zation of encryption and decryption is proposed. Computer simulations confirmed its possibility.

A Decode-and-Forward Scheme for LDPC Coded Three-Way Relay Fading Channels

In this paper a low-density pairwise check(LDPC) coded three-way relay system is considered, where three user nodes desire to exchange messages with the help of one relay node. Since physical-layer network coding is applied, two time slots are sufficient for one round information exchange. In this paper, we present a decode-and-forward(DF) scheme based on joint LDPC decoding for three-way relay channels, where relay decoder partially decodes the network code rather than fully decodes all the user messages. Simulation results show that the new DF scheme considerably outperforms other common schemes in three-way relay fading channels.

In-Network Puncturing for Delay-Efficient Rate Control of Distributed Video Coding in Wireless Video Sensor Networks

Wyner-Ziv Video Coding （WZVC） is considered as a promising video coding scheme for Wireless Video Sensor Networks （WVSNs） due to its high compression efficiency and error resilience functionalities, as well as its low encoding complex- ity. To achieve a good Rate-Distortion （R-D） per- formance, the current WZVC paradi^prls usually a- dopt an end-to-end rate control scheme in which the decoder repeatedly requests the additional deco- ding data from the encoder for decoding Wyner-Ziv frames. Therefore, the waiting time of the additional decoding data is especially long in multihop WVSNs. In this paper, we propose a novel pro- gressive in-network rate control scheme for WZVC. The proposed in-network puncturing-based rate control scheme transfers the partial channel codes puncturing task from the encoder to the relay nodes. Then, the decoder can request the addition- al decoding data from the relay nodes instead of the encoder, and the total waiting time for deco- ding Wyner-Ziv frames is reduced consequently. Simulation results validate the proposed rate con- trol scheme.

FAST ALGORITHM FOR REED-SOLOMON CODE AND THE HARDWARE IMPLEMENTATION

This paper described a signal processor for the Reed-Solomon (R-S) code using micro-programming. For the purpose of fast encoding and decoding,a formula for detecting two errors is derived, and the Qian search method for the decoding process is improved. The number of searches is significantly reduced from 256 to 4. At the same time, the circuit is simplified and the speed is increased. For the convenience of programming, a micro-programming compiling package is developed. The package can be used for programming of different formats of R-S code signal processors of DAB, MD and DCC. The software and the hardware can be used for error correcting, error detecting and error compensation of different formats of R-S code.

Simple optical decryption based on a modified joint transform correlator technique

A new encryption/decryption system for optical information security is proposed in this paper. We used an iterative Fourier transform algorithm to optimize the encrypted hologram as well as the decryption key as phase-only elements. The optical decryption was implemented by superimposing the encrypted hologram and the decryyption key in a simple optical setup. Numerical simulation and optical experiment have confirmed the proposed technique as a simple and easy implementation for optical decryption, demonstrating potential applications in optical information security verification.

WIRELESS FLOPPY DISK DRIVE USING TURBO CODEC IN HOME AND OFFICE WIRELESS LAN

A wireless floppy disk drive is proposed for home and office wireless local area network. The proposed floppy drive uses turbo coding-decoding in the transceiver PHYsical layer (PHY) to improve system performance. The data transmission rate is about 11 Mbps. The covered distance is 30 meters in obstructed environment and 100 m in free space.

LOW-COMPLEXITY REDUNDANCY INSERTION IN MULTIPLE DESCRIPTION CODING

Multiple description coding has recently been proposed as a joint source and channel coding to solve the problem of robust image transmission over unreliable network, and it can offer a variety of tradeoff between signal redundancy and transmission robustness. In this letter, a novel pre- and post-processing method with flexible redundancy insertion for polyphase downsampling multiple de- scription coding is presented. The proposed method can be implemented as pre- and post-processing to all standards for image and video communications, with obvious advantages. Simulation results show that this approach can reduce the computational complexity while provide a flexible redundancy in- sertion to make the system robust for any packet loss situation over different networks.

A Concatenated ML Decoder for ST/SFBC-OFDM Systems in Double Selective Fading Channels

This paper presented a concatenated maximum-likelihood (ML) decoder for space-time/space-frequency block coded orthogonal frequency diversion multiplexing (ST/SFBC-OFDM) systems in double selective fading channels. The proposed decoder first detects space-time or space-frequency codeword elements separately. Then, according to the coarsely estimated codeword elements, the ML decoding is performed in a smaller constellation element set to searching final codeword. It is proved that the proposed decoder has optimal performances if and only if subchannels are constant during a codeword interval. The simulation results show that the performances of proposed decoder is close to that of the optimal ML decoder in severe Doppler and delay spread channels. However, the complexity of proposed decoder is much lower than that of the optimal ML decoder.

THE UNIFICATION OF DIFFERENTIAL ENCODING AND DECODING ALGORITHM FOR VARIOUS SIGNALS

Many monographs point out that differential encoding and decoding is necessary for ef- fectual information transmission against phase ambiguity while seldom discuss the reason why phase ambiguity will emerge inevitably.Available algorithms are specially designed for certain modulation scheme;these algorithms cannot satisfy the requirement of soft-defined radio,which perhaps demands a uniform algorithm for different modulations.This paper proposes a new opinion on phase ambiguity from the view of probability.This opinion believes that modulating symbol sequence can affect,at optimum sampling epoch,the modulated waveform as oscillating carrier has done,and so the stochastic sequence leads to phase ambiguity.Based on a general signal model,this paper also puts forward a novel universal algorithm,which is suitable for different signals,even some new ones,by configuring several parameters.

Design of bilayer lengthened LDPC codes over expanded graph for relay channels

This paper presents a low complexity optimized algorithm for design of bilayer lengthened LDPC(BL-LDPC) code for decode-and-forward relay system.The design is performed over the expanded graph of the BL-LDPC code,which consists of the original bilayer graph and the extra added relaygenerated parity check bits.To build up our proposed optimized algorithm,we present a modified Gaussian approximation algorithm for the expanded structure of the BL-LDPC code.Then using the proposed optimized algorithm,we find the optimum overall expanded graph of the BL-LDPC code.Simulation results show that the BL-LDPC codes obtained by our proposed optimized algorithm have excellent bit-error-rate performances and small gaps between the convergence thresholds and the theoretical limits when transmitted over the additive white Gaussian noise channels.

APPLICATION OF TURBO CODES IN HIGH-SPEED REAL-TIME CHANNEL

The time delay of Turbo codes due to its iterative decoding is the main bottleneck of its application in real-time channel. However, the time delay can be greatly shortened through the adoption of parallel decod-ing algorithm, dividing the received bits into several sub-blocks and processing in parallel. This letter mainly discusses the applicability of turbo codes in high-speed real-time channel through the study of a parallel turbo decoding algorithm based on 3GPP-proposed turbo encoder and interleaver in various channel. Simulation re-sult shows that, by choosing an appropriate sub-block length, the time delay can be obviously shortened with-out degrading the performance and increasing hardware complexity, and furthermore indicates the applicability of Turbo codes in high-speed real-time channel.

Screen image sequence compression method utilizing adaptive block size coding and hierarchical GOP structure

To compress screen image sequence in real-time remote and interactive applications,a novel compression method is proposed.The proposed method is named as CABHG.CABHG employs hybrid coding schemes that consist of intra-frame and inter-frame coding modes.The intra-frame coding is a rate-distortion optimized adaptive block size that can be also used for the compression of a single screen image.The inter-frame coding utilizes hierarchical group of pictures(GOP) structure to improve system performance during random accesses and fast-backward scans.Experimental results demonstrate that the proposed CABHG method has approximately 47%-48% higher compression ratio and 46%-53% lower CPU utilization than professional screen image sequence codecs such as TechSmith Ensharpen codec and Sorenson 3 codec.Compared with general video codecs such as H.264 codec,XviD MPEG-4 codec and Apple's Animation codec,CABHG also shows 87%-88% higher compression ratio and 64%-81% lower CPU utilization than these general video codecs.

New Applications of EXIT Chart for Iterative Decoding Systems

Applications on iterative control and multiple input multiple output (MIMO) system were developed. Two new charts derived from extrinsic information transfer (EXIT) chart were employed as the designing tools, which are called as output mutual information chart with defined iterative degree (DID) chart and near optimum output mutual information (NOMI) chart respectively. Different from the EXIT chart, they can show the iterative performance on the whole signal-to-noise ratio range with one single curve, whereas computation complexity is greatly reduced compared with conventional bit error ratio (BER) performance curve. The iterative control was implemented according to a near-optimum iterative degree vector determined by NOMI chart, the reasonability of uncertain parameters was analyzed in one MIMO system. The concepts were illustrated based on bit-interleaved coded modulation with iterative decoding (BICM-ID).

Side information refinement in distributed video coding

Side information （SI） is one of the key issues in distributed video coding （DVC） and affects the compression performance of DVC largely. This paper proposes an SI refinement algorithm, in which the Wyner-Ziv （WZ） frame is split into two parts based on checkerboard pattern, and the two parts are encoded independently but decoded sequentially. In the decoding process, the part 1 is first decoded with the initial SI and partially decoded part （PDP） 1 is used to improve the motion vectors （MVs） and SI of both parts. At the next stage, the part 2 is decoded with the improved SI and PDP 2 is used to further refine MVs of the part 2. Then, SI of both parts are further refined. Simulation results show that the proposed algorithm can improve the peak signal to noise ratio （PSNR） by up to 1.43 dB when compared with traditional DVC codec.