一种改进的综合纹理和形状特征的图像检索方法  

Image Retrieval Based on Combined Color with Texture Feature

作　　者：刘黎宁[1] 侯榆青[1] 高士瑞[1] 

机构地区：[1]西北大学信息科学与技术学院,西安710127

出　　处：《小型微型计算机系统》2012年第5期1141-1144,共4页Journal of Chinese Computer Systems

基　　金：陕西省自然科学基础研究计划项目(2009JM8004)资助;西北大学研究生自主创新项目(08YZZ34)资助

摘　　要：提出一种利用小波进行综合纹理和形状特征的具有旋转、平移和尺度不变性的图像检索算法.使用角向矩加权方向定义图像的主方向来进行坐标轴的旋转矫正,得到图像的旋转不变性表示;采用具有平移和尺度不变性的小波变换对图像进行小波分解,利用各子带的能量作为纹理特征;利用小波分解的逼近子图重构图像并进一步利用Hu不变矩提取其形状特征.最后对纹理和形状特征进行高斯归一化,综合其特征进行检索.实验中对算法的尺度不变性、旋转不变性、平移不变性及对噪声的不敏感性进行了验证,实验结果证明了该算法具有更高的鲁棒性和查准率.A new method for image retrieval combining texture with shape features using wavelet is proposed,which has the property of rotation,shift and scale invariance.The rotation-normalized image is achieved by rotating the principal axis to the weighted orientation of angular moments.The adopted wavelet,which has the property of shift and scale invariance,is used to decompose the orientation-normalized image.The wavelet energy of every sub frequency band is extracted to be the texture feature of the image.Using the sub-graph rebuild the graph and then extract the shape features by Hu invariant moments.Gaussian model is used to normalize the texture and shape features.Finally,the feature combined texture with shape is used to retrieve.The experimental results show that the proposed algorithm has better robustness and precision.

关 键 词：图像检索 小波变换 纹理特征 形状特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]