一种综合PHOG形状和小波金字塔能量分布特征的图像检索方法  

An Image Retrieval Method Synthesizing PHOG Shape Description and Wavelet Pyramid Energy Distribution

作　　者：袁杰[1] 魏宝刚[1] 王李冬[1,2] 

机构地区：[1]浙江大学计算机科学与技术学院,浙江杭州310027 [2]杭州师范大学,浙江杭州310012

出　　处：《电子学报》2011年第9期2114-2119,共6页Acta Electronica Sinica

基　　金：国家自然科学基金项目(No.60673088)

摘　　要：形状是物体的一个重要属性,在图像检索中发挥着重要作用.PHOG梯度方向直方图金字塔是最近出现的一种表达力较强的形状特征,但其对自然背景图像的检索效果不佳.图像的小波分解能得到图像能量场在各频域的分布,从而可用于图像检索.本文提出一种新的小波金字塔能量分布特征,在基于SVM分类器的检索框架下,与图像的PHOG形状描述特征动态结合,对图像进行检索.在Corel图像集上的检索试验表明,本文提出的综合方法比基于PHOG形状描述特征和小波金字塔能量分布特征两种方法有更好的效果.Shape is a very important attribute of objects,and it plays an important role in image retrieval.Pyramid of histogram of orientation gradients（PHOG） is a powerful expressive description of object shapes.But it has a poor performance in searching images with a nature background.The distribution of image energy field among different frequency domains can be obtained by wavelet decomposition,so the wavelet character can be used in image retrieval.In this paper,a new wavelet pyramid energy distribution character is extracted from images,and is dynamically combined with PHOG character in an image retrieval framework based on SVM classifier.Experiments on a Corel image subset show that our method has a better effect than methods only using PHOG shape description character or wavelet pyramid energy distribution character.

关 键 词：PHOG 小波分析 形状特征 SVM分类器 图像检索 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]