基于浮点坐标系的图像处理方法及应用研究

发布时间：2018-06-20 09:42

本文选题：图像处理 + 图像变形　；参考：《浙江工业大学》2016年博士论文

【摘要】：现实生活中的图像在人类感知领域中所传达的信息比其他方式更真切、直观,因此利用图像进行目标识别一直都是信息智能系统最热门的课题之一。从最初的模型匹配到现在的三维图像识别,人们一直孜孜不倦地从硬件和算法上寻找一种彻底解决机器模仿人类认知的模型。“一种高维度的通用目标识别建模策略”是一种解决图像识别技术的新的构想。针对目前图像目标识别过程中无法彻底克服的姿态、表情和光照影响,本文在图像文件的格式采用一种新的描述图像对象的集合和集合间的相互关系的图像文件格式,即浮点坐标系下的图像数据模型。浮点坐标系首先与高维空间几何学的“HDG”文件相一致,标准化了仿生模式识别中的数据格式;最主要的是它将图像文件的对象信息充分挖掘出来,使得文件数据从整齐的排列中解脱出来,不仅可以按照图像内容任意组合,而且使得图像的变形、联接运算变得更加直观、简便、生动。本文的研究工作和创新点主要有:1.本文构建了一种新的图像文件的浮点坐标系格式,并对浮点坐标系的文件格式做了详细介绍。借鉴常用的图像文件格式,并结合高维空间几何学数据的基本格式,本文构建了一种用在图像文件处理领域的图像数据格式:将每个像素点的信息“R,G,B”扩展到“AX,AY,R,G,B”;为了体现图像的立体性,可以将图像内容分层,各层之间通过联接点进行组合联接。与现有图像文件格式以像素值为体现形式的数据模型不同,浮点坐标系的文件格式同时使用了像素点的位置信息和像素值信息。2.网格是图像处理的一种常用模型,本文在浮点坐标系的基础上,构建了三角网格模型。在每个三角网格中,三个顶点作为坐标约束顶点,组成三角形坐标,其余的点可以通过两个三角坐标值确定。本文推导了三角形坐标表达与笛卡儿坐标系的相互转换数学模型,并论证了三角形坐标表达的一系列性质:仿射不变性、连续性、可分解性、线性不变性和旋转、缩放、平移不变性。这些性质在图像处理领域的应用中起到了重要作用。浮点坐标系与三角形坐标将图像的描述方法从繁琐而且死板的像素点中挣脱出来,转变到以图像内容为基础的网格坐标模型上来,实现了图像的结构和纹理分离,使图像处理更加自由3.利用三角形坐标的线性不变性和旋转、缩放、平移不变性,本文提出了一种基于三角坐标的图像校正归一化算法。首先给出三角形坐标下的归一化算法;然后通过实验,对有姿态、缩放、表情影响的人脸图像进行校正、归一、去表情化处理。4.利用三角形坐标的变形连续性,本文提出了一种基于三角形坐标对小样本图像进行插值、重构图像的方法。得到的插值图像有效细化了学习样本,从而改进了仿生模式识别算法中的有效覆盖问题,实验结果证明改进后的算法是有效的。5.提出了一种描述表情的矢量坐标模型。本文通过图像在三角形坐标下的特征点结构变化,得到7种表情的位移矢量和表情矢量坐标模型,并通过实验证明了同种表情的方向性,验证了表情矢量坐标模型的有效性。最后利用表情矢量坐标模型对人脸表情进行识别研究,实验效果良好。
[Abstract]:In real life, the information conveyed in the field of human perception is more authentic and intuitive than other ways. Therefore, image recognition has always been one of the most popular topics in the information intelligent system. From the original model matching to the present 3D image recognition, people have been searching for the hardware and algorithms tirelessly. A high dimension universal target recognition modeling strategy is a new idea to solve the image recognition technology. In view of the attitude, expression and illumination that can't be completely overcome in the process of image target recognition, this paper uses a new description in the format of image file. The image file format of the relationship between the set and the set of the image object, that is, the image data model in the floating point coordinate system. The floating point coordinate system is first consistent with the "HDG" file of the high-dimensional space geometry, and the data format in the bionic pattern recognition is standardized. The most important is that it fully excavates the object information of the image file. In order to remove the file data from a neat arrangement, it can not only be combined arbitrarily according to the content of the image, but also make the image deformed. The connection operation becomes more intuitive, simple and vivid. The research work and innovation of this paper are mainly as follows: 1. the floating point coordinate system of a new image file is constructed and the floating point coordinates are constructed in this paper. The file format of the system is introduced in detail. Using the common image file format for reference, and combining the basic format of high dimensional space geometry data, this paper constructs an image data format used in the image file processing field: the information "R, G, B" of each pixel point to "AX, AY, R, G, B", in order to embody the stereoscopic image of the image, It is different from the data model of the existing image file format in the form of pixel value. The file format of the floating point coordinate system uses the pixel location information and the pixel value information.2. grid is a common model in the image processing. This paper is in floating point coordinates. On the basis of the system, a triangular mesh model is built. In each triangular mesh, three vertices are used as the vertices of coordinate constraints, and the other points can be determined by two triangular coordinates. This paper derives the mathematical model of the mutual conversion between the triangular coordinate expression and the Cartesian coordinate system, and demonstrates the expression of the triangle coordinates. A series of properties: affine invariance, continuity, decomposability, linear invariance, rotation, scaling and translation invariance. These properties play an important role in the application of image processing. The floating point coordinate system and triangular coordinate transform the image description method from the tedious and rigid pixels to the image. On the basis of the grid coordinate model, the structure and texture of the image are separated. The image processing is more free and 3. uses the linear invariance and rotation, scaling and translation invariance of the triangle coordinates. In this paper, a kind of image correction normalization algorithm based on triangular coordinates is proposed. First, the normalization of the triangle coordinates is given. And then, through experiments, the face images which have the influence of attitude, zoom and expression are corrected, and then, to deal with the deformation continuity of.4. using triangular coordinates, a method of interpolation and reconstruction of small sample images based on triangular coordinates is proposed. The effective coverage problem in the bionic pattern recognition algorithm is improved. The experimental results show that the improved algorithm is an effective.5. to propose a vector coordinate model describing the expression. In this paper, the displacement vector and the expression vector coordinate model of 7 kinds of expressions are obtained by the change of the feature point structure in the triangle coordinates, and the experiment is carried out through the experiment. The directionality of the same expression is proved, and the validity of the expression vector coordinate model is verified. Finally, the facial expression is identified by the expression vector coordinate model, and the experimental results are good.
【学位授予单位】：浙江工业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】