基于改进傅里叶描绘子的CBIR系统  

CBIR SYSTEM BASED ON IMPROVED FOURIER DESCRIPTORS

作　　者：彭地[1] 彭源[2] 

机构地区：[1]复旦大学计算机科学技术学院,上海200433 [2]上海电力学院计算机与信息工程学院,上海200090

出　　处：《计算机应用与软件》2011年第9期89-92,共4页Computer Applications and Software

基　　金：上海高校选拔培养优秀青所教师科研专项基金(z-2008-14)

摘　　要：在现有的基于傅里叶描绘子的CBIR系统中,为了提高检索速度,一般需要舍去物体轮廓经傅里叶变换后的大部分高频分量。当物体轮廓在细节部分具有较高能量时,此方法不具备有效性。为尽可能保证检索准确率并兼顾检索速度,在原有傅里叶描绘子上进行扩展,避免直接舍去高频分量,引入Fisher判别分析法将描绘子映射到子空间进行降维,并保证了映射后子空间的类内样本距离最小化和类间样本距离最大化。基于此改进傅里叶描绘子的CBIR系统具有更高的准确率与检索速度。In present CBIR（Content based image retrieval） systems based on Fourier Descriptors,in order to speed up the retrieval,it is generally required to omit the majority of high frequency sub values by Fourier Transform from the object contour.But such method is ineffective when there is stronger energy in the detailed parts of the object contour.In order to guarantee maximum retrieval precision as well as retrieval speed,the original Fourier descriptors are extended to avoid the direct omission of high frequency sub values by introducing Fisher discriminant analysis to map the descriptors to the subspace for dimension reduction,in addition assuring the minimization of distances between samples within a class as well as the maximization of distances between samples of different classes of the subspace after mapping.CBIR systems based on the improved Fourier descriptors bear higher precision rate and retrieval speed.

关 键 词：CBIR基于内容的图像检索 傅里叶描绘子 傅里叶变换 Fisher判别分析法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]