基于特征点几何关系的约束的近似图像检索-计算机应用技术专业论文.docx

JIIIII JIIIII II I II III IIllll llI IIIJ Y2048855 浙江人学硕J：学位论文 摘要 摘要 随着多媒体技术及因特网的快速发展，因特网上图像数据来源不断扩大，图 像的数量增长迅速。相对于文字信息，图像表达更加生动、易于理解，因此各行 各业对图像的使用越来越普及，比如医学图像、卫星遥感图像、新闻图像、动漫 图像等等。在图像给我们工作和生活带来便利的同时，我们也面临一个问题，如 何高效地从海量图像数据中查找到用户需要的数据，因此图像检索技术成为关注 的重点。 本文分析现有基于图像内容的近似图像检索方法的优缺点，深入研究基于空 间金字塔的近似图像检索技术。主要工作包括利用特征点之间几何约束关系保持 了图像中结构性特征以及空间对齐金字塔匹配方法保持了图像尺度无关属性的 特点，提出了一种新颖的近似图像检索方法，并在实验中验证这一方法提高了近 似图像检索的准确率。接下来为了进一步研究近似图像技术，我们提出基于 Laplacian金字塔模型的图像检索技术，通过提取图像的边缘信息，逐层匹配，构 建图像金字塔，最后融合多层次的图像匹配结果，提高近似图像技术的准确率。 关键词：近似图像检索，几何关系，空间金字塔，Laplacian金字塔 浙江人学硕I：学位论文 浙江人学硕I：学位论文 Abstract Abstract With the rapid development of multimedia technology and intemet，the source of image on the intemet continues to expand．The number of images has grown rapidly． Compared with text，images can describe things in more vivid ways，which make it easy to understand，SO images is becoming increasing popular in every sectoL such as medical images，satellite remote sensing images，news images，anime images，and SO on．Although images make our lives a lot easier,we have to face a problem how to efficiently retrieve the image that we need from the mass．That is why image retrieval technology is very hot． This thesis analyzes the advantages and disadvantages of existing content-based image retrieval methods，and study the pyramid-based imaging technology．In this thesis we proposed a novel approach of the retrieval of Near Duplicate Images(NDI) by introducing the geometrical structures of visual words into Spatially Aligned Pyramid Matching which Can robustly handle spatial and scale variants．The image retrieval experiments conducted in this work show that Our approach outperforms SPM and SAPM significantly．Then we propose a image retrieval method which build the image pyramid according to Laplacian pyramid method，Then integrate multi—level image matching results together to improve accuracy rate． Keywords：Near Duplicate Images retrieval，Geometrical structures，Spatial Pyramid， Laplacian Pyramid 浙江人学硕I：学位论文 浙江人学硕I