图像模型

This work deals with the depth information estimation of 3-D surface in a series of images from different shot angle.

解决从不同角度拍摄的一系列平面图像中解算图像中物体的景深问题，使用一个新颖的二阶先验模型来建模光滑3D物体的表面信息，此模型比传统的一阶模型有很大优势。

A human model using polynomial convolution surface is presented which is modeled by articulated skeletons convolved with some kernel function. The convolution surface and curve correspondence theorem under orthogonal projection is presented. The human contour is approximated with a convolution curve and 3D human pose can be estimated using a nonlinear optimization algorithm.

首先建立了一种新颖的多项式密度卷积曲面人体模型，该模型由连接体线骨架与某种卷积核卷积而成，整个曲面是一张连续曲面，调节多项式和半径参数可以让人体任意变形；给出了正交投影下卷积曲面和曲线对应的概念和相关定理，为二维的图像轮廓和三维人体模型之间建立对应关系；根据正交投影下卷积曲面与曲线对应定理，用卷积曲线去逼近人体的图像轮廓曲线，从而估计和恢复出三维人体运动姿态。

We integrate the advantage of semantic depiction using ontology and the advantage of complicated targets image modeling using Bag Of Words model, complicated targets se mantic reasoning model is introduced using the ontology concept as a random variable in Bag Of Words model. This model can map the ontology concept to the image.

结合本体技术语义表达的优势和Bag Of Words模型的复杂目标图像建模的优势，将复杂目标本体中的本体概念作为随机变量引入到Bag Of Words模型中，提出了基于本体的复杂目标语义推理模型，建立了本体概念同图像的映射关系。

Mumford and Shah＇s variational model in the 2-phase piecewise constant case is very efficient in image segmentation.

Mumford—Shah两相分片常数模型是一个有效的图像分割模型，但当模型用于带有噪声的图像时，其水平集解法存在对初始解和长度参数敏感这两个问题。

By the minimum risk Bayes decision theory, this paper develops a new way of image segmentation: establish the mathematics model of image segmentation; estimate the probability density of grey scales and figure out its math-expectation and square difference that accord with normal distribution and the loss function; and judge the every pixel dot in the image according to the minimum risk Bayes decision theory and determine whether it is of target or non-target images, thereby realizing the extraction of the target image.

依据最小风险贝叶斯决策理论，提出了一种基于最小风险贝叶斯决策的图像分割方法。首先建立图像分割的最小风险贝叶斯决策模型，对灰度级类条件概率密度估计出其符合正态分布的数学期望和方差以及损失函数，再依据最小风险贝叶斯决策理论对图像中的每一像素点进行目标图像和非目标图像的类别判断，从而实现目标图像的提取。

A new self-adaptive background approximating and updating algorithm based on optical flow theory is first presented in this paper.

在光流场等技术的基础上，提出了一种自适应背景逼近更新方法，以适应光线的变化以及噪声的影响；根据彩色差值模型计算当前帧图像和当前逼近背景图像的差分图像；并引入Gauss模型实现了运动目标的自适应阈值分割。

Based on the two face images of the same person at front and half-profile view and an MPEG4 3D generic face model, we estimate the poses of the face at these two images and establish their 2D mesh models. Multiple views of the same face can be synthesized by image interpolation and warping, meanwhile occlusion problems are solved reasonably.

基于同一人脸的正面和半侧面图像，引入三维通用人脸模型估计其初始朝向，分别建立两图像的二维网格模型；然后采用图像插值和变形技术合成不同朝向人脸，并合理地解决遮挡问题。

In this paper,the storage forms of sound and image information are first discussed.Then,the authors study and analyse methods and technology that can be used in sound and image retrieval,and give classified expositions.Meanwhile,this paper makes comparative research on data models of the multimedia information retrieval system,proposes that the design of MIRS may base on a RDB system,and it can support storage and retrieval of sound and image information by expanding its ability of managing complex objects and multimedia data.

本文首先分析了声音与图像信息的存贮处理，接着分类阐述了声音、图像信息检索可以采用的方法和技术；在此基础上，对多媒体信息检索系统的数据模型进行了比较研究，提出在目前技术条件下，多媒体信息检索系统的设计可采用现有成熟的关系数据库系统作为基础，并在关系模型里增加处理复杂对象与多媒体数据的能力，以实现对声音、图像等多媒信息的存贮与检索。

According to the function of distributed emulator, optical information processing and image multi-object tracking system, the following three aspects were emphasized in this dissertation. Firstly, a vast virtual terrain, ground radar models and canyons were created for the distributed emulator with modeling software Creator. The radars were driven to scan during emulating and the real time terrain code was sent to Radar module. Secondly, an optics channel simulating the optics camera of a virtual aircraft was implemented with simulating software Vega. The 2-D information got from the channel was translated and sent to multi-sensor information fusion system as an optics information source. Thirdly, the dissertation studied some motion estimation methods and analyzed the factors which influence the speed and the precision of the block matching algorithms. Then an improved block matching method was proposed for image multi-object tracking system and applied in the system combined with frame difference method.

论文根据多传感信息融合实验系统中的分布式飞行仿真系统、光学信号处理器和图像多目标跟踪系统三个子系统的功能要求，具体完成了以下三方面的工作：首先利用建模软件Creator创建分布式飞行仿真系统中的大面积地形、场景中的地面雷达模型和峡谷地形，并在仿真过程中实时驱动雷达模型的扫描，同时为后续雷达仿真模块提供实时地形编码；其次利用仿真软件Vega，在分布式仿真系统中为光学信号处理器设计实现了光学通道，模拟飞行器的光学摄像机的功能，并且对由此通道获得的目标二维光学信息进行转化处理，作为光学信息源发送给后续的多传感信息融合器；最后针对图像多目标跟踪系统，在研究图像分割运动估计算法的基础上，通过分析影响块匹配算法速度和精度的因素，改进了块匹配运动估计算法，结合差分法可应用到目标跟踪系统中。

A content based adaptive mesh generation algorithm is proposed in this paper. At first, a watershed algorithm of mathematical morphology is used to segment image into many homogeneous regions so that watershed lines can better represent image structure, contours, and edges. Boundaries of these regions are approximated by polygons that can be represented with a few parameters. Each end point of polygonal sides serves as candidate node of mesh model.

文中提出的基于图像内容的自适应网格模型生成算法，首先利用数学形态学中的水线算法把编码图像分割成许多纹理一致的区域，所分割的区域反映了图像的结构、轮廓和边界；再对这些区域的边界进行多边形拟合，得到多边形各个边的端点作为网格模型的节点，以这些节点为基础就能生成一个 Delaunay 三角形网格。

The absorption and distribution of chromium were studied in ryeusing nutrient culture technique and pot experiment.

采用不同浓度K2CrO4(0，0.4，0.8和1.2 mmol/L)的Hoagland营养液处理黑麦幼苗，测定铬在黑麦体内的亚细胞分布、铬化学形态及不同部位的积累。

By analyzing theory foundation of mathematical morphology in the digital image processing, researching morphology arithmetic of the binary Image, discussing two basic forms for the least structure element: dilation and erosion.

通过分析数学形态学在图像中的理论基础，研究二值图像的形态分析算法，探讨最小结构元素的两种基本形态：膨胀和腐蚀；分析了数学形态学复杂算法的基本原理，把数学形态学的部分并行处理理念引入到家实际应用中。