一种基于形状特征的图像检索方法

A More Distinctive Shape Descriptor For Image Retrieval

基于内容的图像检索技术中，形状特征是最重要的图像特征部分。傅立叶描述子（FD），因为它高效性，稳定性，低计算复杂度和少量参数运行的优点而被广泛应用。目前已经存在的大多数傅立叶描述子往往直接使用它的傅立叶系数，本文我们应用主成分分析Principal Component Analysis（PCA）方法在傅立叶描述子中，从而提出基于PCA的傅立叶描述子的形状描述和检索方法。本论文的实验采用MPEG-7标准图像数据库来比较检索性能。实验结果表明PCA-FD的方法实现了更短时间内的更好检索精确率。

Shape is one of the most important image features in Content Based Image Retrieval （CBIR）. Many image classification and retrieval applications are based on the shape descriptors. So far, Many shape descriptors has been developed. Among the developed descriptors, Fourier Descriptor （FD） is well known because of its efficiency, stability, low computational cost and few parameter tuning. FD directly uses its Fourier coefficients, but in this paper, we apply Principal Component Analysis （PCA） on FD and propose a PCA-based FD for shape representation and image retrieval. PCA based FD do not use Fourier coefficients directly, instead, we describe a shape by applying PCA to FD with a large number of Fourier coefficients. The experiment is based on the standard MPEG-7 shape database to compare the retrieval performance. The experiment shows that PCA-FD achieved better retrieval precision and lower time cost.