• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文学位 >A framework for color image indexing and object recognition: Spectral and spectral spatial gradients.
【24h】

A framework for color image indexing and object recognition: Spectral and spectral spatial gradients.

机译：彩色图像索引和对象识别的框架：光谱和光谱空间梯度。

获取原文
获取原文并翻译 | 示例
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

This dissertation addresses the effects of illumination pose (direction) and color on the problems of object recognition and image indexing, and presents color models as well as algorithms to overcome these effects. Most of the color recognition/indexing approaches concentrate on illumination color invariance in order to improve the utility of color recognition. However, effects caused by illumination pose on three-dimensional object surfaces and locally varying illumination have not received notable attention. The work presented in this dissertation is one of the first approaches that have been recently developed to account for illumination pose in addition to illumination color. Furthermore, it addresses global and local changes in illumination color. Our unique approach to various illumination conditions provides a framework with three different levels of color processing for recognition/indexing. Three color spaces are proposed within which to create object feature descriptions invariant to illumination pose and color. The first is a chromaticity space formed by computing spectral derivatives of the logarithmic image irradiance. This Spectral Gradient (SG) space provides invariance to illumination pose and global changes of illumination color. However, if the spectral qualities of the illumination vary spatially within an image, object descriptions in the SG space may be distorted beyond usefulness. Taking the spatial derivative of SG information creates Spectral Spatial Gradients (SSGs). This SSG space is invariant to illumination pose, as well as global and local changes in illumination color. While SSGs have a greater degree of invariance than SGs, they may have a reduced discrimination power. It is well known that most invariant features in computer vision result in decreased discrimination power. Nonetheless, the combination of SG and SSG information can be used to extract local illumination color change. Removing the effect of the extracted illumination from the SG information creates Adapted Spectral Gradient (ASG) information. This space maintains both the discrimination power of SGs and the invariance property of SSGs. These three proposed color spaces create a unified framework with a rich potential for local and global descriptors of object color, thus furthering the advancement of color recognition approaches.
机译：本文着眼于照明姿态（方向）和颜色对物体识别和图像索引问题的影响，并提出了颜色模型以及克服这些影响的算法。大多数颜色识别/索引方法都集中在照明颜色不变性上，以提高颜色识别的效用。然而，由照明对三维物体表面造成的影响以及局部变化的照明并未引起人们的关注。本文提出的工作是最近开发的用于解决照明颜色和照明姿势的首批方法之一。此外，它解决了照明颜色的整体和局部变化。我们针对各种照明条件的独特方法为识别/索引提供了具有三种不同级别的色彩处理的框架。提出了三个颜色空间，可在其中创建不依赖于照明姿势和颜色的对象特征描述。第一个是通过计算对数图像辐照度的光谱导数形成的色度空间。该光谱梯度（SG）空间为照明姿势和照明颜色的全局变化提供了不变性。但是，如果照明的光谱质量在图像内在空间上变化，则SG空间中的对象描述可能会超出使用范围而失真。采用SG信息的空间导数将创建频谱空间梯度（SSG）。该SSG空间对于照明姿势以及照明颜色的全局和局部变化是不变的。尽管SSG的不变性比SG大，但它们的辨别力可能会降低。众所周知，计算机视觉中的大多数不变特征会导致辨别力降低。但是，SG和SSG信息的组合可以用于提取局部照明颜色变化。从SG信息中删除提取的照明效果会创建自适应光谱梯度（ASG）信息。这个空间既保持了SG的区分能力，又保持了SSG的不变性。提出的这三种颜色空间创建了一个统一的框架，具有丰富的潜力，可用于对象颜色的局部和全局描述符，从而进一步促进了颜色识别方法的发展。

著录项

  • 作者

    Berwick, Daniel Obadiah.;

  • 作者单位

    University of Michigan.;

  • 授予单位 University of Michigan.;
  • 学科 Computer Science.
  • 学位 Ph.D.
  • 年度 2001
  • 页码 95 p.
  • 总页数 95
  • 原文格式 PDF
  • 正文语种 eng
  • 中图分类 自动化技术、计算机技术;
  • 关键词

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号