摘要
在微软图像库基础上提出了一种新的基于傅里叶算法的图像检索方法。主要是综合了hu不变矩和二维极坐标傅里叶描述子对图像提取特征向量。相似性度量采用加权的欧氏距离算法。Hu不变矩可以对物体的整体性进行描述,平移、旋转和尺度变化都不能影响提取的特征向量。二维极坐标傅里叶描述子可以对形状的相似性进行定量判别,具有较强的通用性。最后,采用线性加权求和的方式计算综合相似度,将具有较高相似度的结果。实验结果表明,该方法可以有效地提高检索的查全率和查准率。
A new image retrieval method is proposed based on Fourier algorithm in Microsoft image library. Improved moment invariants and two-dimensional polarFourier descriptors are comprehensively used to describe the shape features. Similarity measure is to use the weighted Euclidean distance algorithm.Hu momentinvariants can describe the shape information of the objects from the whole aspects, with the in-variance of translation, rotation and scale. Two-dimensional polar Fourier descriptive clauses can shape a quantitative analysis of the comparability, strong commonality. Finally, using linear weighted summation of the way of calculating the integrated similarity, has high similarity of the results. Experimental results show that this method can effectively im- prove retrieval of recall and Drecision.
关键词
图像
形状特征
二维极坐标傅里叶描述子
检索
Image Shape feature Two-dimensional polar Fourier de-scriptor Retrieval
