中文摘要：

为了更好地了解变换域方法在图像检索中的检索效果，比较分析了Zernike矩、GFD、PCET、RCFT和RFMT等现有的5种基于变换域的形状描述方法．分别在不变性、计算复杂性、噪声鲁棒性、有效特征数的选取方面对其进行了深入的分析和比较．选用MPEG7 CE Shape-1PartB中的1400幅图像构成的图像库对这些方法进行检索和性能测试，并实际应用于由500幅病例图构成的图像库的医学图像检索．在研究噪声影响时，对各测试集图片加上不同程度的高斯噪声．通过比较分析及实验结果验证，Zernike矩和GFD方法的检索性能最好，有良好的抗噪性，因而适合于医学图像检索的实际应用．

英文摘要：

In order to better understand the effect of transform domain method in image retrieval, we give a comparison of the five transform-based shape descriptors, which are Zernike moments, gener- ic Fourier descriptor（ GFD）, polar complex exponential transform（ PCET）, Radon chamfer Fourier transform（RCFT） and Radon Fourier-Mellin transform（RFMT） were compared. Besides, analyses on invariance, computation complexity, noise robustness and effective number of features were con-ducted in detail. The five descriptors were tested by using 1 400 images from MPEG7 CE Shape-1 Part B. Furthermore, these descriptors were practically used for the medical image database inclu- ding 500 images. To study the effect of the noise on retrieval, the Gaussian noise was applied to the two databases. It can be seen from the comparative analysis and the experimental results that the per- formance of Zernike moments and GFD are the best. They have robust noise immunity and are suit- able for the practical application of medical image retrieval.