基于选择性多分辨率Kohonen网络的自适应灰度图像分割方法

针对多阈值分割中存在的问题,提出一种新的基于测量空间的空间聚类方法,即基于选择性多分辨率Kohonen网络的自适应灰度图像分割方法.算法的优点在于不需预知图像分割级数,而能够动态地根据图像局部特性决定.实验证明该方法具有良好的适应性.     

一种基于目标特征的多门限图像分割方法

本文提出了一种基于目标特征的多门限图像分割方法。它通过对特征量匹配的方式将属于景物的先验空间知识引入分割过程,使图像分割由单纯的灰度聚类过程变成灰度－几何空间聚类过程。实验结果表明,本文提出的方法可以完整地分割出复杂背景图像中本身具有多个灰度层次的景物。     

基于层次聚类的多颜色空间融合分割方法

针对自然图像分割中,由于单一的颜色空间难以表示复杂多变的场景信息以及目标与背景的低对比度等产生的过分割和误分割问题,提出了选择RGB、XYZ和LUV这3个颜色空间的增强图像进行基于层次聚类的融合分割的新方法.对Berkeley分割图像库中的多幅图像进行了多组分割实验,并与Mean-shift等多种经典分割方法进行了定性...     

一种基于Markov过程和证据理论的多源图像融合分割方法

文章从多传感器图像的属性研究出发，提出图像分割的新概念。建立图像邻域系统的Markov过程和基于Markov过程的Dempster-Shafer融合分割模型。并充分利用象素间的空间邻接关系。用证据理论描述图象分类的不确定性，较好地解决了传感器图像分割信息不全的问题，实现了景物图像的准确分割。     

基于直方图极大值准则可变参数图像阈值分割方法

提出了一种基于直方图极大值准更则图像阈值分割方法，通过调节参数可对多类图像进行分割。阉值分割是实时图像处理首选的分割算法。本文通过一维灰度直方图确定极大值，从而获得图像中多个目标区域和背景区域阉值，分割出多个区域，将每个区域赋为不同的灰度级，根据具体所要分割的目标再进行分割。实验证明该方法分割效果较好，分割速度快，实时性较好。     

多分辨率图像锥的FCM聚类图像分割算法

模糊C-均值（FCM）聚类算法是一种基于像素分类的图像分割方法，在分割的过程中，仅仅利用了像素点的灰度信息，但在灰度密度丰富变化和图像的对比度不明显的情况下，物体和背景的分布将相互重叠而密不可分，往往得不到满意的分割效果。为了解决上述问题，现提出了一种基于多分辨率图像锥的模糊C-均值聚类图像分割算法。该方法利用多分辨技术产生多分辨率图像锥，将图像从空间信息引入，考虑图像的局部特性，使分割算法局限于图像的子图像中，物体和背景比单纯运用FCM更容易区分，且算法稳定性高，速度快。     

基于自适应多尺度与轮廓梯度的遥感图像分割网络

遥感图像分割算法易受环境因素干扰，如物体遮挡、光照不均匀等。现有的深度学习遥感图像语义分割方法通常采取端到端的编解码结构，但针对相似度较高物体的结构和轮廓，仍存在分割不准确的问题。为了提高算法鲁棒性、分类准确率，提出一种基于轮廓梯度学习的深度卷积神经网络遥感图像语义分割算法。为了提高预测特征图的质量，首先基于SegNet模型，提出自适应注意力的多通道多尺度特征融合网络（D-MMA Net），其中D-MA block采用基于注意力的自适应多尺度模块，根据学习到的权重自适应地对不同尺度特征进行提取，以获得更多有效的高级语义特征。为进一步细化提取物体的边界，基于Sobel边缘检测算子原理提出可学习的轮廓提取模块。最后将轮廓信息与多尺度语义特征相结合，以增强对图像空间分辨率的鲁棒性。实验结果表明，所提算法提高分割的准确率，对于不规则物体边界，能有良好的分割效果。     

基于模糊聚类算法的体育运动视频图像分析应用

针对当前的体育运动视频图像分割方法存在图像分割结果粗糙以及空间畸变率高等缺陷,提出基于模糊聚类算法的体育运动视频图像分割方法。介绍了模糊聚类算法的基本理论,借助时域差分图像建立含有正态分布和灰度值的二阶模糊属性,赋予模糊属性S隶属度函数,进而对时域差分图像进行模糊聚类,再通过边缘检测得到运动视频图像分割结果。实验结果证明,该方法的空间精度高、噪声迭代性能好、空间畸变率低,可精准分割复杂运动视频图像,得到高清画面。     

SAU-Net:融合压缩注意力机制的多器官图像分割

为了对头颈部多器官CT图像实现精准分割,减少放射治疗对人体正常组织的损伤,本文提出一种基于卷积神经网络的图像分割方法——SAU-Net算法,该算法基于加入残差连接的3D U-Net实现.针对器官尺寸差异较大而引起的分割精度不准确问题,引入压缩注意力模块,通过非局部的空间注意力机制增加对全局特征的编码能力,聚合多尺度上下...     

基于免疫算法的交通标志分割

针对现有道路交通标志图像分割算法中实时性和鲁棒性较差的问题,提出一种基于免疫算法和RGB颜色空间的交通标志图像分割新方法。该方法先用RGB颜色空间将交通标志从实景图中定位出来,利用免疫算法快速搜索的机理,对已定位的图像进行阈值T寻优,并用该阈值T分割图像。实验结果表明,该算法与基于遗传算法的图象分割算法相比较,具有较好的鲁棒性和实时性,且适用于不同光照条件下的交通标志图像分割。     

Multi-objective data collecting strategies for wireless sensor network based on the time variable multi-salesman problem and genetic algorithm

Comparing to the traditional data collecting method with data route,the technology of wireless mobile nodes has gradually became a new technique in the wireless sensor network.As the solution to the visiting order of the static nodes was an intrinsic NP-hard problem,a more general multi-objective data colleting strategies based on multi-mobile nodes was proposed.The proposed data collecting technique was abstracted as a model of time variable multiple traveling salesman problem.Belonging to a discrete optimal problem,the proposed model was solved by with a proposed hybrid genetic algorithm to determine the paths of the multi-mobile nodes.The convergence analysis of the proposed algorithm was given.With the experiment of open dataset,the proposed model based on the time variable multiple traveling salesman problem and the proposed hybrid genetic algorithm certify a certain improvement to the efficiency and real-time ability.     

Double Gaussian mixture model for image segmentation with spatial relationships

In this paper, we present a finite mixture model based on a Gaussian distribution for image segmentation. There are four advantages to the proposed model. First, compared with the standard Gaussian mixture model (GMM), the proposed model effectively incorporates spatially relationships between the pixels using a Markov random field (MRF). Second, the proposed model is similar to GMM, but has a simple representation and is easier to implement than some existing models based on MRF. Third, the contextual mixing proportion of the proposed model is explicitly modelled as a probabilistic vector and can be obtained directly during the inference process. Finally, the expectation maximization algorithm and gradient descent approach are used to maximize the log-likelihood function and infer the unknown parameters of the proposed model. The performance of the proposed model at image segmentation is compared with some state-of-the-art models on various synthetic noisy grayscale images and real-world color images.     

New fractional-order LADRC scheme based on a novel filtered-Bode’s ideal transfer function for integer-order systems

In this paper, a new Fractional-Order Linear Active Disturbance Rejection Control scheme (FO-LADRC) is proposed to enhance the robustness against loop gain variations of the standard Active Disturbance Rejection Control (ADRC) in the case of uncertain integer-order systems. A new filtered Bode’s ideal transfer function (F-BITF) is proposed to be used as a reference model in the design approach of the proposed control scheme to ensure the dynamic behavior of the closed-loop BITF to the controlled system. A Fractional-order Extended State Observer (F-ESO) is used in the proposed FO-LADRC structure to approximate the system to be controlled by a filtered fractional-order integrator. The fractional order of the F-ESO is a design parameter to tune to achieve the desired overshoot of the closed-loop step response. For the tuning of FO-LADRC structure, an analytical method is proposed. The performance of the proposed FO-LADRC and the Chen’s et al. FO-ADRC structures are evaluated thorough numerical simulation, and then validated in practice in the case of a Cart-Pendulum. Both the simulation and the experimental results show that the proposed FO-LADRC is able to achieve the desired dynamics of the F-BITF and guarantee the robustness with respect to the controller gain variation and the system parameter uncertainties. The comparative study conducted also reveals that the proposed control scheme is more robust than that of Chen.     

New equalisation approach for maximising data rate in discrete multi tone systems

This article presents a new time domain equaliser (TEQ), which can be used in discrete multi-tone (DMT) systems to maximise the bit rate. The proposed TEQ is tested in the fast Fourier transform based discrete multi-tone (FFT–DMT) system, and in a proposed discrete cosine transform based discrete multi-tone (DCT–DMT) system. The objective of the proposed DCT–DMT system is to make use of the energy compaction property of the DCT to reduce the channel effects on the transmitted signals. A mathematical model of the proposed TEQ is presented. Simulation experiments have been carried out to test the effect of the proposed TEQ with the FFT–DMT system and the proposed DCT–DMT system. The results show that the performance of the DCT–DMT system with the proposed TEQ is better than the FFT–DMT system with this TEQ over the eight standard carrier serving area (CSA). The results also show that employing the proposed TEQ in the DCT–DMT system can achieve a high bit rate, ranging from 2.899?Mbps to 5.369 Mbps.     

基于双曲线特征的浅地层探地雷达杂波抑制与合成孔径成像研究

本文根据浅地层探地雷达图像的特点,提出了一种基于图像处理技术的杂波抑制方法。并在此基础上,进而提出了一种实现探地雷达合成孔径成像的快速方法。通过对实测数据进行处理,结果表明,所提杂波抑制方法取得了很好的效果,且通用性强;所提的快速合成孔径成像方法与普通合成孔径成像方法相比,所得图像的分辨力相当,且所提方法减少了杂波成分的影响,使处理速度得到了极大的提高,成像的效果更好。     

用于CMOS图像传感器的12位低功耗单斜坡模数转换器设计

 本文提出了一种应用于CMOS图像传感器中的高精度低功耗单斜坡模数转换器(single slope analog-to-digital converter)设计方案.该ADC方案由可变增益放大器、前置预放大器和动态锁存比较器组成.相比现有的设计方案,本文提出的电路在不牺牲噪声性能的前提下,具有更低的功耗和更小的芯片面积.通过集成列并行的单斜坡模数转换器在最新设计的高精度高速CMOS图像传感器设计中,实验结果证明了设计的有效性.     

基于深度学习的快速QTMT划分

与之前的编码标准相比,多功能视频编码(versatile video coding,VVC)进一步提高了压缩效率。嵌套多类树的四叉树(quadtree with nested multi-type tree,QTMT)结构是提高编码增益的关键之一,同时极大地增加了编码复杂度。为降低VVC编码复杂度,提出了一种基于深度学习的快速QTMT划分方法。首先,提出了注意力?非对称卷积结构来预测划分模式的概率。然后,基于阈值提出了快速划分模式决策。最后,提出了编码性能与时间的代价函数来求解最优阈值,提出了阈值决策方法。实验表明,算法在不同档次下的时间节省分别为48.62%、52.93%、62.01%,BDBR分别为1.05%、1.33%、2.38%。结果表明,算法的时间节省和率失真性能优于其他快速算法。     

Novel Architecture for Efficient Implementation of Dimmable VPPM in VLC Lightings

In this paper, a new architecture is proposed to achieve complexity efficiency in implementing variable pulse position modulation (VPPM). VPPM, specified in IEEE 802.15.7, can support wireless communication and dimming control simultaneously using visible light. The proposed architecture is based on the VPPM signal property in which the transition point of the modulated output is obtained by counting the sample index and comparing it to both the assigned dimming factor and the transmitting data. Therefore, the proposed architecture can be composed of simple logics, including a counter, a comparator, and an inverter, all of which are insensitive to the dimming resolution in contrast to a conventional codeword‐table method. This paper describes the verification of the proposed algorithm through a register‐transfer level implementation of the codeword and proposed architectures. In comparison with the codeword‐table method, the proposed method gains a nine‐fold complexity reduction at a 1% dimming‐step resolution.     

Novel video stabilization for real-time optical character recognition applications

This paper presents a novel video stabilization algorithm for real-time optical character recognition (OCR) applications. The proposed method generates output frames in order to stabilize the position of a target word that will be recognized by the OCR application. Unlike in conventional algorithms, in the proposed algorithm, a causal low pass filter is not applied to the trajectory of the target word for reducing the high frequency component of camera motion. The proposed algorithm directly calculates the stable position of the word using two forces: the force used to pull the target word to the center of an output frame and a back force used to return the center of an output frame to the center of an input frame. Hence, the proposed algorithm significantly minimizes the time take to respond to sudden camera movement. Although the proposed method may not outperform state-of-the-art video stabilization in terms of video stability, the proposed technique is much more appropriate for real-time OCR applications than the conventional techniques in terms of accuracy, computational cost, processing delay, and the time taken to respond to sudden camera movement. Simulation results prove the superiority of the proposed method over conventional techniques for real-time OCR applications.