Texture segmentation using independent-scale component-wise Riemannian-covariance Gaussian mixture model in KL measure based multi-scale nonlinear structure tensor space

This paper proposes a novel texture segmentation approach using independent-scale component-wise Riemannian-covariance Gaussian mixture model (ICRGMM) in Kullback-Leibler (KL) measure based multi-scale nonlinear structure tensor (MSNST) space. We use the independent-scale distribution and full-covariance structure to replace the covariant-scale distribution and 1D-variance structure used in our previous research. To construct the optimal full-covariance structure, we define the full-covariance on KL, Euclidean, log-Euclidean, and Riemannian gradient mappings, and compare their performances. The comparison experiments demonstrate that the Riemannian gradient mapping leads to its optimum properties over other choices when constructing the full-covariance. To estimate and update the statistical parameters more accurately, the component-wise expectation-maximization for mixtures (CEM²) algorithm is proposed instead of the originally used K-means algorithm. The superiority of the proposed ICRGMM has been demonstrated based on texture clustering and Graph Cuts based texture segmentation using a large number of synthesis texture images and real natural scene textured images, and further analyzed in terms of error ratio and modified F-measure, respectively.     

粗糙集理论中基于属性重要性的离散化方法

连续属性离散化一直是机器学习领域中亟待解决的关键问题之一。目前已有许多离散化方法,存在的主要问题是断点集的选取带有很大的主观性,导致大多数的离散化算法难以得到较满意的离散效果。分析了当前研究中常用的离散化方法,提出了一种基于属性重要性的离散化方法,利用遗传算法,把最小断点集作为优化目标,利用遗传算法,以属性重要性构造适应度函数,保证了原决策系统的不可分辨关系,避免了决策表信息的损失,所得结果相对来说比较客观。     

Functionality of cow beef in coarse and fine ground model systems

Patties and sausage batters were made from trimmed brisket, chuck, striploin, hindshank, silverside, thick flank and rib trim from grass-fed Friesian cows. Functionality of each meat cut was determined from cook yield and true stress and strain of cooked sausage batters and cook yield, dimensional changes and peak force of cooked patties. Colour of raw and cooked slices from the meat cuts and cooked batters was evaluated using a colorimeter and a consumer panel. Composition of raw meat from all cuts was similar; chuck and rib trim had higher pH values and were more red than meat from other cuts. Cooked batters from striploin, chuck and hindshank had higher stress and strain values than gels from other cuts. Patties made from thick flank and striploin were the softest and had the highest cook yield. Batters and patties made from meat with higher salt soluble protein content retained more water and fat on cooking, were more cohesive and had higher bind strengths. Manufacturers can use functionality characteristics to select meat for specific products.     

基于粗集理论的数据离散化方法

对后继阶段的机器学习或数据挖掘过程而言，决策系统中连续属性值的离散化具有非常重要的意义。本文系统地研究了基于粗集理论的数据离散化方法：提出一种计算候选断点集合的算法；定义概念“选择概率”来合理、有效地度量和区分候选断点的相对重要性；最后基于这一概念提出一种确定结果断点子集的启发式算法,理论分析及仿真结果表明，算法的综舍性能优于文献报道的同类算法．     

基于特征点跟踪的视频序列平面提取算法

平面结构是一种最为常见和重要的几何特征,平面提取在很多领域有着重要的应用。提出了一种毋须恢复摄像机三维信息、直接从视频序列中提取出平面的算法。先通过在视频相邻帧之间建立极线约束下的射影变换模型,来计算出由平面诱导的单应,而每一个单应都对应着场景中的一个平面;然后根据计算出的单应对视频的每一帧进行分割,进一步提取出视频中的平面结构。一系列真实视频序列的实验验证了本算法的有效性和稳定性。     

肤色和手势识别方法研究

随着信息技术的快速发展,手势输入比于其他人机交互方式更直接。然而,利用手势或人的姿态来控制机器之前,必须先从图像画面中切割出正确的肤色目标。通过对图像切割技术的研究,给出一种结合概率神经网络用于图像分割中得出最佳切割阈值的想法,以达到切割复杂背景中不同的皮肤颜色。     

Automatic objects segmentation with RGB-D cameras

Automatic object segmentation is a fundamentally difficult problem due to issues such as shadow, lighting, and semantic gaps. Edges play a critical role in object segmentation; however, it is almost impossible for the computer to know which edges correspond to object boundaries and which are caused by internal texture discontinuities. Active 3-D cameras, which provide streams of depth and RGB frames, are poised to become inexpensive and widespread. The depth discontinuities provide useful information for identifying object boundaries, which makes automatic object segmentation possible. However, the depth frames are extremely noisy. Also, the depth and RGB information often lose synchronization when the object is moving fast, due to different response time of the RGB and depth sensors. We show how to use the combined depth and RGB information to mitigate these problems and produce an accurate silhouette of the object. On a large dataset (24 objects with 1500 images), we provide both qualitative and quantitative evidences that our proposed techniques are effective.     

Relationships between carcass quality parameters and genetic types

It is important for the meat industry, including pig farmers and breeding companies, to know the composition of the carcass including the proportions of the different commercial cuts when making decisions on the type of pigs to be produced and marketed. Carcass composition is influenced significantly by the genotype of the animals. The aim of this work was to characterise carcasses from five different pig genotypes, by means of the quantification and comparison of their physical composition. Carcasses from 500 gilts from five different genetic types were studied. These lines were based on the following breeds: Large White, Landrace, Duroc, Piétrain and a Meishan synthetic population developed from a cross with a Large White based line. Measurements were taken in the carcasses directly with a ruler and with the Fat-O-Meat'er. The carcasses were cut following the European reference method and the four main joints were dissected. Carcasses from the Piétrain based line, which was halothane negative, presented the highest killing-out (83.34%) and were the shortest (81.81cm). The Piétrain based line was also the leanest and the Meishan based line the fattest. The highest proportion of ham (270.9g/kg) and the lowest proportion of belly (97.97g/kg) were found in the Piétrain line. The proportion of lean in all of the dissected cuts was higher in carcasses from this line while the Meishan line presented the highest proportions of intermuscular fat in all of the pieces. Carcasses came from the Piétrain line received the highest conformation scores and they were leaner and with better ham yield.     

基于图切割和密度聚类的视频行人检测算法*

现有视频行人检测方法把行人检测看成一个有监督的两类(即行人和背景)学习问题,区分视频中的行人和背景,并不能很好解决行人的姿态变化和行人间的遮挡问题.文中提出基于图切割和密度聚类的行人检测算法,把行人检测看成一个多类的无监督学习过程.在训练阶段,首先对每个训练样本计算多级梯度方向直方图-局部二分模式(HOG-LBP)特征,然后对多级HOG-LBP特征所属的每个图像块分配不同的权值.为了区别行人的不同部位并赋权值,采用基于图像块的图分割方法从背景中分割行人所在的图像块.最后,再采用基于密度峰值的聚类算法对正样本和负样本分别进行无监督的聚类.在测试阶段,首先通过计算样本特征与每个聚类中心的距离,然后使用前5个最短距离进行投票,判断其是否包含行人.实验证明,文中算法较好解决行人的姿态变化和行人间的遮挡问题,并且随着训练样本的增加,能取得和目前最优行人检测方法可比较的结果.     

一种基于Graph Cuts的SAR图像分割方法

在最小化由马尔科夫随机场(MRF)图像分割模型建立的能量函数方面,基于Graph Cuts的alpha-expansion是一种比较有效的算法.但是,由此算法构建的s/t图中边的数目非常多,运算速度很慢.为了减少alpha-expansion算法的计算量,本文在标号为alpha的像素向其它像素膨胀的过程中,先隔离非alpha类间的联系,而只考虑alpha类与非alpha类之间的关系,从而避免了alpha-expansion算法需要构造辅助结点的问题,减少了s/t图中边的数目,提高了算法的计算效率.因放松了非alpha类间的关系对alpha膨胀的约束,使得算法可以更容易得跳出能量函数的局部极小点而获得更优的分割结果.实验中将改进的算法与传统的基于Graph Cuts的算法做了对比,显示了新算法在运算时间和最小化能量方面的有效性.