一种基于活动围道的纹理图像分割方法

本文将Gabor滤波器和各向异性扩散方程相结合，提出了一种基于活动围道的无监督纹理图像分割算法。采用基于总变分流的扩散函数，各向异性扩散方程可以有效地在保留纹理图像大尺度边界信息的同时对图像纹理区域进行平滑，获得比原始图像更易分割的简化图像。但是平滑过程中纹理信息的丧失，限制了该方法的通用性和有效性。为了在利用各向异性扩散方法的同时有效地提取和利用纹理信息，我们利用Gabor滤波器提取一组表征纹理方向性和尺度性的特征图像，同时将原始图像作为表征纹理灰度信息的一个特征通道考虑。再利用矢量形式的各向异性扩散方程对特征图像进行边界保持的各向异性平滑。我们将基于区域灰度统计参数估计的活动围道分割方法扩展到矢量空间，来对平滑后的纹理特征量进行分割。实验证明利用该纹理分割算法可以获得较好的效果。     

基于灰度和梯度的视网膜下积液分割

视网膜下积液的容积计算可以被用来评价不同类型的视网膜和黄斑紊乱,因此准确找到下积液的轮廓也就十分必要。但是,目前关于视网膜下积液边界分割的研究很少。本文的目的是根据图像的灰度和梯度特征来分割出视网膜下积液,该方法利用图论分割算法找到经常出现下积液的区域边界,限定搜索区域后利用自适应阈值和梯度阈值来达到分割的目的。实验表明,该方法对于分割视网膜下积液效果良好。     

改进CV模型在宫颈细胞图像分割中的应用

宫颈细胞图像分割中边缘信息十分重要。将边缘信息融入传统CV模型,提出新的几何模型,结合同质区域与边缘信息的新模型弥补了CV模型只采用全局灰度信息的不足,边缘函数采用负指数形式使得细胞边界分割结果更加准确。实验结果表明:新模型能克服传统CV模型缺点,并减少迭代次数。     

基于高频增强和区域标定的涡轮叶片缺陷定位

提出一种基于高频增强和区域标定的涡轮叶片缺陷定位检测算法,设计涡轮叶片缺陷定位检测工作原理,采用高频滤波的方法来锐化叶片图像,采用预处理的办法来初步控制过度分割现象,由灰度图像生成梯度图像,构建一个噪声抑制算子,解决过度分割问题,进行了涡轮叶片的缺陷特征模型构建和缺陷信息计算,采用高频增强算法和区域标记的方法,减少由于噪声和纹理细节产生的伪边界,提高缺陷定位性能。仿真实验表明,采用该算法能有效提高涡轮叶片的缺陷定位检测性能,突出损伤边界,较好地抑制噪声。具有较高的工程应用价值。     

基于改进自动语义标注的图像识别方法

当前图像识别采取边缘算子切割技术,图像目标边缘会出现一些灰度急剧变化的情况.采用阈值法,图像中存在的阴影,照度不均匀,各处的对比度不同的情况,所以不能采用全局阈值法,而采用局部阈值法,即将图像划分为若干个小图像,先对各子图像用阈值法进行分割,再将分割后的小区域合并在一起,得到整幅图像的完整分割结果,这样,不同的区域由于不同的情况,就可以选取相应的阈值,达到最好得分割效果.     

基于SVM的花岗岩薄片图像石英分割方法研究

花岗岩薄片不同矿物成分在图像中可能存在相同灰度分布甚至相似纹理。通过结合正交偏光镜镜下图片干涉色信息和单偏光镜图片信息分析,提高了花岗岩薄片石英分割的准确率。方法是从单偏光镜和正交偏光镜下图片中提取图像块的特征信息,运用支持向量机分类方法对图像块进行分类,在分类为石英的图像块中选取种子,再根据两张图片信息,用区域生长方法完成石英分割。该方法实现了自动化分割,分割准确率高,实验证明该方法切实可行。     

基于横噪声消除的医学粘连图像边缘分割算法

针对CCD型医学图像中,在细胞粘连区域信号附近出现的较强的横条纹噪声干扰,影响图像信息的正确性,分割后会存在边界模糊和锯齿条文的问题,为提高医学图像分割效果,提出了一种基于横条纹噪声消除的医学粘连图像边缘分割算法。分析了医学图像中边沿横条纹噪声的原因,通过wold纹理模型与多尺度马尔可夫随机场模型,利用确定性随机场和不确定性随机场的谱属性不同的特征,将医学图像边沿的干扰特征分离开,有利于对粘连医学图像进行分割。实验证明,方法有效地去除了横条纹噪声并很好地保留了图像的边缘和细节信息,同时运算复杂度低。     

车牌识别系统中车牌定位与字符分割方法的研究

根据车牌区域的灰度分布特征,提出一种定位车牌区域并进行字符分割的新方法。该方法先利用车牌区域灰度分布特征,对车辆图片进行粗定位并得到车牌上下边界;再根据车牌区域字符紧密性特征,利用垂直投影方法初步得到车牌的左右边界;然后根据字符宽度和间距进一步修正车牌的左右边界,最后根据字符宽度和间距的比例进行字符分割。经实验证明,该方法具有在复杂背景下适应性好、抗干扰、实时性好、定位精确等特点。     

改进的ASM灰度图像目标定位技术

首先介绍灰度模型的目标定位算法,在该算法基础上提出了一种改进的算法,将主动形状模型的灰度图像中目标定位算法应用到视频图像中,同时针对处理结果进行科学评价。实验结果证明了本文提出算法的可行性和高效性。     

远程微小目标图像的粘连区域高精度分割模型

远程微小目标图像粘连区域存在孔洞和震荡突变问题,使得传统针对微小图像粘连区域分割模型,容易产生过分割现象,分割效率较低,获取的分割结果存在较大的偏差,提出了一种优化马尔科夫随机场(MRF)的分割方法,塑造远程微小目标图像优化分割模型,采用改进势函数获取MRF的条件概率,处理图像的粘连分割问题,全面分析图像分割区域的相对高度和面积,通过优化的MRF标记准确分割区域和过分割区域,运算过分割区域同邻域的紧密度,采集紧密度最大的领域并与之融合,实现远程微小目标图像粘连区域的高精度分割。实验结果说明,所提模型可对远程微小目标图像的粘连区域进行准确分割,准确率为97%。     

Semantic categorization of digital home photo using photographic region templates

In this paper, a semantic categorization method in generic home photos is proposed. In recent years, the semantic categorization of image has been a challenge due to the proliferation of digital home photos. Our approach is to detect semantically meaningful concepts contained in a photo. The proposed method incorporates an intermediate level of concepts, called local concept, so that it catches well semantic meaning of local regions of image as bridging the semantic gap of the low-level features and high-level category concepts. To detect the local concepts from the home photo, region segmentation by photographic region template and concept merging is also proposed. The efficacy of the proposed semantic categorization method was demonstrated with 3828 general home photos. The experiment results showed the proposed categorization method would be useful to detect multiple semantic meaning of the home photos.     

Text segmentation: A topic modeling perspective

In this paper, the task of text segmentation is approached from a topic modeling perspective. We investigate the use of two unsupervised topic models, latent Dirichlet allocation (LDA) and multinomial mixture (MM), to segment a text into semantically coherent parts. The proposed topic model based approaches consistently outperform a standard baseline method on several datasets. A major benefit of the proposed LDA based approach is that along with the segment boundaries, it outputs the topic distribution associated with each segment. This information is of potential use in applications such as segment retrieval and discourse analysis. However, the proposed approaches, especially the LDA based method, have high computational requirements. Based on an analysis of the dynamic programming (DP) algorithm typically used for segmentation, we suggest a modification to DP that dramatically speeds up the process with no loss in performance. The proposed modification to the DP algorithm is not specific to the topic models only; it is applicable to all the algorithms that use DP for the task of text segmentation.     

Copyright protection of multiple CT images using Octonion Krawtchouk moments and grey Wolf optimizer

This paper proposes a novel Octonion Krawtchouk Moments (OKMs) transform to deal with a set of images in a compact manner, and based on this transform, a local zero-watermarking scheme is proposed to protect the copyright of CT medical images. The scheme first annotates regions of interest (ROIs) on seven medical images and then uses the OKMs of these ROIs to construct a single feature image called zero-watermark. This scheme adopts the gray Wolf Optimizer (GWO) algorithm to have a blind nature and to improve robustness against common image processing manipulations and attacks (zero-watermarking requirements). In addition, our scheme uses the trained U-net (R231) model to reduce the search space for the GWO algorithm and prevent this algorithm from getting stuck in a local optimal solution. The experimental results show that the proposed method is very robust against common image processing manupilations and attacks and has superiority compared with superb other zero-watermarking methods.     

基于模糊C-均值聚类图像分割技术研究

主要研究了均值聚类图像分割问题。针对传统的聚类图像分割算法对图像地分割精度较低等问题,提出一种基于模糊控制的C-均值聚类快速图像分割新方法。本文采用快速模糊C-均值聚类算法对图像分割。实验结果表明,图像分割边缘清晰,分割效果明显优于传统的聚类图像分割算法。     

基于小波域高斯-柯西混合模型的SAR图像降噪声算法

(Symmetric alpha-Stable)模型是小波域一种精确的图像模型 ，但是其计算复杂性高,本文在分析 模型的基础上，提出了高斯-柯西模型，将此模型作为图像的小波域先验模型信息，并用Bayesian估计器，算法复杂性上有显著的降低。另外，图像对数变换后的统计特性发生变化 ，需要在变换过程中加入均值调整的过程。实验结果表明，本文的算法模型在较好的保持了SAR图像结构的基础上，能较好的保留图像边缘信息和纹理特征，并能有效的抑制图像的斑点噪声，取得了良好的效果。     

钢板表面缺陷深度主动学习高效分类方法

针对钢板表面缺陷图像分类传统深度学习算法中需要大量标签数据的问题,提出一种基于主动学习的高效分类方法。该方法包含一个轻量级的卷积神经网络和一个基于不确定性的主动学习样本筛选策略。神经网络采用简化的convolutional base进行特征提取,然后用全局池化层替换掉传统密集连接分类器中的隐藏层来减轻过拟合。为了更好的衡量模型对未标签图像样本所属类别的不确定性,首先将未标签图像样本传入到用标签图像样本训练好的模型,得到模型对每一个未标签样本关于标签的概率分布(probability distribution over classes, PDC),然后用此模型对标签样本进行预测并得到模型对每个标签的平均PDC。将两类分布的KL-divergence值作为不确定性指标来筛选未标签图像进行人工标注。根据在NEU-CLS开源缺陷数据集上的对比实验,该方法可以通过44%的标签数据实现97%的准确率,极大降低标注成本。     

Two dimensional quaternion valued singular spectrum analysis with application to image denoising

Compared to the traditional single color plane based image denoising methods, the quaternion valued singular value decomposition (QSVD) exploits the relationship among different color planes. Hence, it has been applied to the color image denoising. On the other hand, compared to the non-overlapping based image denoising methods, the two dimensional real valued singular spectrum analysis (2DSSA) constructs the trajectory matrix with many elements in the matrix being overlapped. Since the 2DSSA exploits the local information within each color plane, it has also been applied to the single color plane based image denoising. However, neither these two image denoising methods can exploit the relationship among the color planes and the local information within each color plane simultaneously. Therefore, this paper proposes a two dimensional quaternion valued singular spectrum analysis (2DQSSA) based method for performing the color image denoising. Our proposed method can enjoy the advantages of both methods. However, the most critical issue for the 2DQSSA is on the selection of these 2DQSSA components. This paper finds that the optimal total number of the 2DQSSA components used for performing the reconstruction is monotonic decreasing with respect to the power of the noise in the image. Therefore, the polynomial fitting approach is proposed to model this relationship. Since the 2DQSSA based denoising method exploits the relationship among the red color plane, the green color plane and the blue color plane, the 2DQSSA based denoising method outperforms the conventional single color plane based denoising methods. Moreover, since the 2DQSSA based denoising method also exploits the local relationship within each color plane, the 2DQSSA based denoising method outperforms the non-overlapping based methods.