2D sparse signal recovery via 2D orthogonal matching pursuit

Recovery algorithms play a key role in compressive sampling (CS).Most of current CS recovery algo-rithms are originally designed for one-dimensional (1D) signal,while many practical signals are two-dimensional (2D).By utilizing 2D separable sampling,2D signal recovery problem can be converted into 1D signal recovery problem so that ordinary 1D recovery algorithms,e.g.orthogonal matching pursuit (OMP),can be applied directly.However,even with 2D separable sampling,the memory usage and complexity at the decoder are still high.This paper develops a novel recovery algorithm called 2D-OMP,which is an extension of 1D-OMP.In the 2D-OMP,each atom in the dictionary is a matrix.At each iteration,the decoder projects the sample matrix onto 2D atoms to select the best matched atom,and then renews the weights for all the already selected atoms via the least squares.We show that 2D-OMP is in fact equivalent to 1D-OMP,but it reduces recovery complexity and memory usage significantly.What’s more important,by utilizing the same methodology used in this paper,one can even obtain higher dimensional OMP (say 3D-OMP,etc.) with ease.     

二维逐步正交匹配追踪算法

在图像等二维信号的应用与处理上,常规压缩感知理论框架存在重构算法效果差、图像块效应明显、对噪声敏感等问题。针对这些问题,根据现有二维观测模型和二维重构算法设计思想,可以设计一种新的重构算法：二维逐步正交匹配追踪算法。该算法借鉴了相关一维重构算法的设计思想,通过每次迭代选取符合阈值条件的多列原子进而正交化处理的步骤,提升了重构效率,改善了恢复图像质量。理论分析和实验结果表明,提出的算法在重构时间得到控制的情况下,得到的图像信噪比有较大提升,超越了现有典型的二维重构算法。     

Accelerating 2D orthogonal matching pursuit algorithm on GPU

Two-dimensional orthogonal matching pursuit (2D-OMP) algorithm is an extension of the one-dimensional OMP (1D-OMP), whose complexity and memory usage are lower than the 1D-OMP when they are applied to 2D sparse signal recovery. However, the major shortcoming of the 2D-OMP still resides in long computing time. To overcome this disadvantage, we develop a novel parallel design strategy of the 2D-OMP algorithm on a graphics processing unit (GPU) in this paper. We first analyze the complexity of the 2D-OMP and point out that the bottlenecks lie in matrix inverse and projection. After adopting the strategy of matrix inverse update whose performance is superior to traditional methods to reduce the complexity of original matrix inverse, projection becomes the most time-consuming module. Hence, a parallel matrix–matrix multiplication leveraging tiling algorithm strategy is launched to accelerate projection computation on GPU. Moreover, a fast matrix–vector multiplication, a parallel reduction algorithm, and some other parallel skills are also exploited to boost the performance of the 2D-OMP further on GPU. In the case of the sensing matrix of size 128 \(\times \) 256 (176 \(\times \) 256, resp.) for a 256 \(\times \) 256 scale image, experimental results show that the parallel 2D-OMP achieves 17 \(\times \) to 41 \(\times \) (24 \(\times \) to 62 \(\times \) , resp.) speedup over the original C code compiled with the O \(_2\) optimization option. Higher speedup would be further obtained with larger-size image recovery.     

基于高阶统计量的压缩宽带频谱盲检测方法

针对认知无线网络中宽带频谱感知受到高速模数转换器(ADC)器件的技术限制,利用压缩感知理论(CS),采用压缩信号处理技术,直接对压缩观测数据进行分析,推导出宽带频谱检测的高阶判决统计量的概率分布特性,并在此基础上提出了一种基于高阶统计量的压缩宽带频谱盲检测算法(HOS-CWSBD).该算法无需任何有关主用户(PU)信号的先验知识、也无需事先重构出原信号就能实现宽带频谱检测.理论分析和仿真结果均表明,与传统的基于压缩感知理论且需要信号重构的压缩频谱感知算法以及基于Nyquist采样数据的非压缩宽带频谱感知算法相比,该算法具有计算复杂度低、感知性能稳定等优点.     

Real-time and low-power processing of 3D direct/inverse discrete cosine transform for low-complexity video codec

The 3D discrete cosine transform and its inverse (3D DCT/IDCT) extend the spatial compression properties of conventional 2D DCT to the spatio-temporal coding of 2D videos. The 3D DCT/IDCT transform is particularly suited for embedded systems needing the low-complexity implementation of both video encoder and decoder, such as mobile terminals with video-communication capabilities. This paper addresses the problem of real-time and low-power 3D DCT/IDCT processing by presenting a context-aware fast transform algorithm and a family of VLSI architectures characterized by different levels of parallelism. Implemented in submicron CMOS technology, the proposed hardware macrocells support the real-time processing of main video formats (up to high definition ones with an input rate of tens of Mpixels/s) with different trade-offs between circuit complexity, power consumption and computational throughput. Voltage scaling and adaptive clock-gating strategies are applied to reduce the power consumption versus the state of the art.     

基于交换-单亲遗传算法的砂岩三维显微图像重建

根据二维显微图像中孔隙的统计特征,重建具有相同特征的三维孔隙结构具有重要意义.本文提出用交换算子产生子代个体,替代传统单亲遗传算法中倒位算子、移位算子和突变算子,构建交换-单亲遗传算法.和传统单亲遗传算法相比,该算法能快速更新子代个体的适应值,算法复杂度大大降低.实验结果表明,与常用的模拟退火重建算法相比,交换-单亲遗传算法能有效加快个体能量下降速度,重建结果的统计特征与二维图像更相符,更适合应用于三维重建问题.     

保角特征结合改进差分进化算法的三维人脸识别

针对三维人脸识别的高复杂度和二维人脸识别无法提供粒状线索的问题,提出一种全自动3D人脸表情识别算法, 该算法主要是提供比2D人脸识别更多的线索,同时降低计算复杂度。首先通过保角映射将3D人脸转化到2D平面,保留了面部变化的线索;然后,提出了基于优化算法的差分进化(DE)算法用于提高识别效率,同时提取最优人脸特征集和分类器参数,加速鲁棒特征(SURF)池描述了所有预期的人脸特征点。在博斯普鲁斯、FRGC v2及自己搜集的人脸数据集上的实验结果表明,本文算法解决了三维人脸识别的高计算复杂度和二维人脸识别的线索低问题,并在不降低识别性能的前提下大大地节约了成本,相比几种较为先进的三维人脸识别算法,本文算法取得了更好的识别效果,有望应用于一些商业人脸识别系统。     

分段迭代匹配追踪图像重构算法

目的 压缩采样匹配追踪（CoSaMP）算法虽然引入回溯的思想,但其原子选择需要大量的观测值且在稀疏度估计不准确时,会降低信号重构精度,增加重构时间,降低重构效率。为提高CoSaMP算法的重构精度,改善算法的重构性能,提出了一种基于广义逆的分段迭代匹配追踪（StIMP）算法。方法 为保证迭代时挑选原子的精确性和快速性,对观测矩阵广义逆化,降低原子库中原子的相干性;原子更新结合正交匹配追踪（OMP）算法筛选原子的准确性与CoSaMP算法的回溯性,将迭代过程分为两个阶段：第1阶段利用OMP算法迭代K/2次;第2阶段以第1阶段OMP算法迭代所得的残差和原子为输入,并采用CoSaMP算法继续迭代,同时改变原子选择标准,从而精确快速地重构出稀疏信号。结果 对于1维的高斯随机信号,无论在不同的稀疏度还是观测值下,相比于OMP、CoSaMP、正则化正交匹配追踪（ROMP）算法和傅里叶类圆环压缩采样匹配追踪（FR-CoSaMP）算法,StIMP算法更加稳健,且具有更高重构成功率;对于2维图像信号,在各个采样率下,StIMP算法的峰值信噪比（PSNR）均高于其他重构算法,在采样率为0.7时,StIMP算法的平均PSNR值比OMP、CoSaMP、ROMP和FR-CoSaMP算法分别高2.14 dB、1.20 dB、3.67 dB和0.90 dB,平均重构时间也较OMP、CoSaMP和FR-CoSaMP算法短。结论 提出了一种改进的重构算法,对1维高斯随机信号和2维图像信号均有更好的重构效率和重构效果,与原算法和现有的主流图像重构方法相比,StIMP算法更具高效性和实用性。     

改进DeepLabV3+的高效语义分割

针对目前高精度的语义分割模型普遍存在计算复杂度高、占用内存大,难以在硬件存储和计算力有限的嵌入式平台部署的问题,从网络的参数量、计算量和性能3个方面综合考虑,提出一种基于改进DeepLabV3+的高效语义分割模型.该模型以MobileNetV2为骨干网络,在空洞空间金字塔池化(AS-PP)模块中并联混合带状池化(MSP...     

Distributed compressed video sensing based on key frame secondary reconstruction

Distributed compressed video sensing scheme combines advantages of compressive sensing and distributed video coding to get better performance, in the meantime, adapts to the limited-resource wireless multimedia sensor network. However, in the conventional distributed compressed video sensing schemes, self-similarity and high sampling rate of the key frame have not been sufficiently utilized, and the overall computational complexity increases with the development of these schemes. To solve the aforementioned problems, we propose a novel distributed compressed video sensing scheme. A new key frame secondary reconstruction scheme is proposed, which further improves the quality of key frame and decreases computational complexity. The key frame’s initial reconstruction value is deeply exploited to assist the key frame secondary reconstruction. Then, a hypotheses set acquisition algorithm based on motion estimation is proposed to improve the quality of hypotheses set by optimizing the searching window under low complexity. Experimental results demonstrate that the overall performance of the proposed scheme outperforms that of the state-of-the-art methods.     

三维图像边缘曲面模型交互式选择抽取算法

提出一种新的三维图像边缘曲面模型选择抽取算法,能以交互方式获取用户感兴趣的三维边缘曲面模型.用户在二维断层图像的切片区选择目标三维边缘曲面模型对应的二维区域,将选定坐标映射到三维区域,利用基于Laplacian算子的三维边缘检测算子检测三维区域内的部分边缘立方体,除噪后置为种子立方体,根据立方体共面和三维区域增长法原理追踪获得用户感兴趣的具有亚体素精度三维图像边缘曲面模型.     

无线传感器网络中APIT--VP三维定位算法

在研究APIT—3D定位算法思想基础上,提出了一种改进的定位算法APIT—VP。新算法解决了APIT—3D算法在节点分布不均匀的情况下定位精度和定位覆盖率较低的问题;在一定程度上避免了PIT—3D测试中出现的OutToIn和InToOut误判错误;并且利用基于中垂面分割法代替原先的网格扫描算法,降低定位运算复杂度,减少能耗。仿真实验结果表明:在无线传感器网络环境理想、300个节点随机部署在100 m×100 m×100 m的三维区域情况下,APIT—VP算法定位覆盖率可达90%,定位误差控制在25%左右,并且与APIT—3D算法相比有效降低了计算复杂度。     

基于正则化弱相关的分布式MWC重构算法

针对实际电磁环境中,信号稀疏度不易准确预知的难题,提出了基于正则化弱相关的分布式调制宽带转换器（Distributed modulated wideband converter, DMWC）重构算法,该算法不依赖稀疏度作为收敛条件。首先将满足弱相关性的原子加入索引集,然后正则化索引集,将新选出的原子加入支撑集。当残差能量达到阈值条件时,停止迭代。最后设置支撑集越界条件,删除支撑集中相关性较小的无效原子,得到最终的支撑集。仿真结果表明,本文算法能大大提高DMWC对信号传输衰减的容忍度。此外,在同等条件下,本文算法的恢复性能优于正交匹配追踪（Orthogonal matching pursuit,OMP）算法。     

量化压缩感知在语音压缩编码中的应用

利用语音信号在离散余弦变换（ DCT）域的近似稀疏性和量化压缩感知理论,文中提出一种基于量化压缩感知的语音压缩编码方案。编码端利用压缩感知技术,将语音信号投影成数据量大大减少的观测序列,然后对观测序列采用Lloyd-Max量化得到量化后的观测样值;解码端直接利用量化后的观测样值,结合重构算法重构出原始语音信号的DCT系数,经过DCT反变换得到重构后的语音信号,并采用后置低通滤波器改善重构语音的听觉效果。该编码方案解码端不需要进行反量化,而是直接利用量化后的观测样值进行重构,有效降低了解码端的运算量及复杂度。仿真结果表明：采用量化迭代硬阈值（QIHT）算法重构效果优于迭代硬阈值算法（IHT）,重构语音的信噪比能达到20 dB以上,MOS分达到3.26。     

AVS-P10音频解码器算法分析与定点化

AVS-P10是我国第一部应用于移动环境的音频编解码国家标准。在分析AVS-P10解码算法的基础上,对参考代码进行了精简和封装。针对定点处理器应用需求,对精简后的代码进行了定点化实现和优化。分别采用CMOS评分与SNR指标,对定点解码器的解码质量进行了主、客观测试,并对优化前后的定点解码器的运算效率进行了比对测试。结果表明,提出的AVS-P10定点解码器的解码音质达到与浮点解码信号的音质相当,且运算复杂度明显下降。     

3D shape acquisition and integral compact representation using optical scanning and enhanced shape parameterization

An efficient computational methodology for shape acquisition, processing and representation is developed. It includes 3D computer vision by applying triangulation and stereo-photogrammetry for high-accuracy 3D shape acquisition. Resulting huge 3D point clouds are successively parameterized into mathematical surfaces to provide for compact data-set representation, yet capturing local details sufficiently. B-spline surfaces are employed as parametric entities in fitting to point clouds resulting from optical 3D scanning. Beyond the linear best-fitting algorithm with control points as fitting variables, an enhanced non-linear procedure is developed. The set of best fitting variables in minimizing the approximation error norm between the parametric surface and the 3D cloud includes the control points coordinates. However, they are augmented by the set of position parameter values which identify the respectively closest matching points on the surface for the points in the cloud. The developed algorithm is demonstrated to be efficient on demanding test cases which encompass sharp edges and slope discontinuities originating from physical damage of the 3D objects or shape complexity.     

CUDA架构下的三维弹性静力学边界元并行计算

针对传统边界元法计算量大、计算效率低的问题,以三维弹性静力学的边界元法为对象,将基于CUDA的GPU并行计算应用到其边界元计算中,提出了基于CUDA架构的GPU并行算法.该算法首先对不同类型的边界元系数积分进行并行性分析,描述了相关的GPU并行算法,然后阐述了边界元方程组的求解方法及其并行策略.实验结果表明,文中算法较传统算法具有显著的加速效果.     

融合稀疏点云补全的3D目标检测算法

基于雷达点云的 3D 目标检测方法有效地解决了 RGB 图像的 2D 目标检测易受光照、天气等因 素影响的问题。但由于雷达的分辨率以及扫描距离等问题,激光雷达采集到的点云往往是稀疏的,这将会影响 3D 目标检测精度。针对这个问题,提出一种融合稀疏点云补全的目标检测算法,采用编码、解码机制构建点 云补全网络,由输入的部分稀疏点云生成完整的密集点云,根据级联解码方式的特性,定义了一个新的复合损 失函数。除了原有的折叠解码阶段的损失之外,还增加了全连接解码阶段存在的损失,以保证解码网络的总体 误差最小,从而使得点云补全网络生成信息更完整的密集点云 Ydetail,并将补全的点云应用到 3D 目标检测任务 中。实验结果表明,该算法能够很好地将 KITTI 数据集中稀疏的汽车点云补全,并且有效地提升目标检测的精 度,特别是针对中等和困难等级的数据效果更佳,提升幅度分别达到 6.81%和 9.29%。     

Computing smooth molecular surfaces

We consider how we set out to formulate a parallel analytical molecular surface algorithm that has expected linear complexity with respect to the total number of atoms in a molecule. To achieve this goal, we avoided computing the complete 3D regular triangulation over the entire set of atoms, a process that takes time O(n log n), where n is the number of atoms in the molecule. We aim to compute and display these surfaces at interactive rates, by taking advantage of advances in computational geometry, making further algorithmic improvements and parallelizing the computations     

基于八叉树结构的三维体素模型检索

随着VR/AR技术发展以及三维模型的广泛应用,实现三维检索具有越来越重要的现实意义.基于模型的检索较好地保留了模型的空间信息和几何特征,其不仅包含模型的表面信息而且还包含模型的内部属性.但是,基于模型的检索往往存在着高存储、高计算的问题.为了解决该问题,本文研究了三维模型预处理及三维模型表示的方法,提出了一种基于八叉树结构的三维体素模型检索方法,即将模型进行体素化处理后提取模型的粗粒度特征和细粒度特征,将两种特征进行融合用八叉树形式表达特征,输入到卷积神经网络中进行训练,最终通过特征的欧氏距离度量实现模型的检索.运用八叉树特征表示法,可以有效地节省体素化存储过程的空间占用量,而且也能保留原始三维网格模型的细节信息.同时考虑到计算性能,本文还在模型体素化的过程中做出一定的改进,通过仅对模型外表面进行体素化,实现了对体素化过程以及数据存储和卷积神经网络训练的优化,大大降低了时间开销.实验中将三维体素模型特征存储在八叉树结构中作为卷积神经网络的输入,结合SOFTMAX代价函数,通过大量的模型训练数据,对该卷积神经网络模型进行训练.与其他同类算法对比,证明了该算法在三维模型检索中的优越性.