【作者】 胡卫军；

【导师】 陈传波；

【作者基本信息】 华中科技大学 ， 计算机应用技术， 2009， 博士

【摘要】 图像表示方法在计算机图形学、图像处理、模式识别、计算机视觉和机器人等研究领域内具有重要的研究意义。非对称逆布局的模式表示模型(NAM)借助于Packing问题的思想，能够有效地表示多种类型模式，是一个通用型的模式表示方法。在NAM模型的基础上，研究了两种新的NAM表示方法，一种是包括多种走向非等腰直角三角形等多种子模式的多子模式表示方法MNAM，另一种是仅包括一种新的多边形子模式的图像表示方法PNAM。并将NAM方法用于图像内容检索，研究了基于NAM的颜色检索和形状特征检索等图像内容检索方法。对NAM中的子模式进行了扩展，研究了点、直线、三角形和矩形等多种典型子模式的MNAM图像表示方法，其中的三角形子模式包括四种方向不等腰的直角三角形。实验结果表明与线性四元树和原有的矩形NAM相比，MNAM方法对图像进行表示时在图像压缩比和重建图像质量方面都有较好的表现。针对MNAM表示，引入了灰度插值的方法，对原有的点、直线、矩形和三角形四种模式重新进行了定义，一个三角形插值块可以通过三个顶点的位置和灰度值按照线性插值的方法生成块内任意一点的灰度值。理论分析和实验结果表明在引入插值方法以后，与MNAM方法相比，算法的复杂度会增加，图像的压缩比可能略有降低，但是图像的重建质量则会得到明显的提高。引入了多边形子模式，对PNAM图像表示方法进行了研究。采用基于区域增长的多边形搜索算法对图像进行编码，对多边形的存储则设计了一种比较精简的形状语法方法。进行图像重建时，则采用一种改进的边标志算法。实验结果表明用PNAM方法对图像进行表示时在图像压缩比和重建图像质量方面都有良好的表现。基于PNAM表示方法对Legendre矩的快速计算进行了研究。PNAM编码以后图像可以看作若干个具有同样灰度值的块，可以利用格林公式将二重积分转化为一重积分，同时又利用当边界被表示为多边形的情况下，线积分还可以直接由顶点的值计算进一步简化矩的计算过程。由于MNAM表示方法可以看作PNAM表示方法的特例，因此最终得到了一个对PNAM和MNAM表示都有效且适合于二值与灰度图像的快速矩计算方法。实验结果表明这种方法在提高矩计算速度上有明显作用。针对当前图像检索中仅限于对关键字进行文字搜索存在的不足出发进行分析，对基于NAM的图像检索方法进行了研究，说明在颜色、形状特征检索方面，基于NAM的图像表示方法都可能有良好的表现。基于NAM的颜色检索算法包括提取图像的灰度直方图和进行直方图匹配两个过程，由于基于NAM的方法将图像表示为一个子模式的队列，该队列与按照象素点表示的图像相比拥有更少的节点数和更小的数据量，因此基于NAM的颜色检索算法具有较低的时空复杂度。通过对多幅二值和灰度图像进行测试，实验数据表明该算法在图像颜色检索上是有效的。基于PNAM的形状图像检索方法，将PNAM表示生成的多边形作为形状提取的对象，采用一种直观、简单的形状特征度量方法，将形状复杂度、半径方差、(?)-直径等多种形状特征用于图像检索，研究了一种有效的形状检索方法。通过对二值和多值图像的实验验证了该方法的有效性。总之，MNAM或PNAM表示方法可以应用于图像表示和图像处理的各个方面，在降低存储空间、加快传输速度、进行图像内容检索等方面具有良好的理论参考意义和实际应用价值。更多还原

【Abstract】 Image representation methods play an important role in computer graphics、imageprocessing、pattern recognition,computer vision and robotics.With the concept ofpacking problem,a non-symmetry and anti-packing pattern representation model(NAM)is presented and the general representation method is given for various kinds of patterns.Based on Non-symmetry Anti-packing pattern representation Model(NAM),twokinds of NAM image representation method are presented,i.e.based Multiple sub-patternNon-symmetry Anti-packing representation Model(MNAM) including Multi-DirectionNon-equilateral right-angled triangle and based Polygonal Non-symmetry Anti-packingrepresentation Model(PNAM).A fast moments compute method based NAM ispresented,firstly apply NAM to the image content retrieval,in addition to giving a imagecomtent retrieval method based on the MNAM and PNAM’s color and shape character.On the basis of analyzing the represent efficiency of NAM,one kind of multiplesub-pattern image representation(MNAM) includeing several typical sun-pattern such asdot,line,triangle and rectangle is defined.Based on the define,the triangle patternincluding four kinds of direction non-equilateral right-angled triangle.the experimentalresults proves that MNAM is better that the LQT and RNAM on image compress andimage rebuilding.Aim to MNAM introduce Interpolation method,redifine the four kinds of pattern,i.e.dot,line,rectangle and triangle.Based on the locations and gray value of threepoints,linear interpolation method can calculate the gray value of any point in the area.The theoretical analysis and the experimental results show that the algorithm complexityis increased,image compression ratio decreased slightly,but the quality of imagereconstruction has a marked improvement after introducing the interpolation method.With respect to the lossless image representation,a novel NAM image representationmethod based on the Polygon subpattern is proposed,which is named PNAM.PNAM applies polygon searching algorithm based on regional growth to encoding image.And arelatively compact shape grammar method is designed for storaging polygon.Animproved algorithm for edge logo is proposed for image reconstruction.Analyzing timeand space complexity of these algorithms and the data of encoded images,Experimentsshow that PNAM has very good performance for image compression ratio and imagequality.A fast calculation method on relatively complex Legendre moment is proposed.Baseon MNAM and PNAM,images after coding can be seen as a number of blocks having thesame gray value.Using Green formula,which replaces integrals on 2-dimensions spaceby integrals on 1-dimension.And When the border is expressed as a polygon,the lineintegral can be directly calculated from the Vertex,which further simplifies thecalculation process on Legendre Moment.Finally,a both effective and suitable fastmoment calculation for binary and gray-scale image is proposed.The experiments showthat that this method plays a significant role in improving the computing speed.The research analyses the limits of image retrieval,which is that the retrievalactually depends on text keyword.So a novel image retrieval method based on NAM isproposed.The novel method could have a good performance at the color and shape-basedretrieval.The algorithm of the NAM for color-based image retrieval is presented and analysisits Complexity.Finally,experiments show that it is effective.The algorithm includes twoprocesses that is extraction of gray-scale histogram and histogram matching.TheNAM-based approach replaces an image by a sub-model queue.The queue has a fewernumber of nodes and a smaller amount of data compared with the pixelimage.Therefore,the proposed approach of image retrieval has a Lower time and spacecomplexity.Through testing a number of binary and gray scale images,the experimentaldata show that the algorithm is effective.A novel and effective approach about image retrieval using PNAM is proposed. Polygon generated by PANM is as the object of shape extraction,This paper presents anew method that is more intuitive and simple than the traditional method.And usingshape complexity、radius Variance and(?) -radius,A effective approach is proposed.Thepaper give the shape retrieval algorithms and analysis their time and space complexity.Finally,The experiments for two-valued and multi-value images show the algorithms areeffective.This method with a combination of the color-based retrieval described in detailin Chapter 5 will generate a better result.In a word,the proposed MNAM and PNAM approaches for the imagerepresentation and manipulations,as envisaged in this paper,shows a very strong promiseand has good potential in business applications dealing with image processing,such asreducing strong room,increasing transmission speed,and improving content based imageretrievel efficiency.更多还原