Abstract: According to the shape characteristics of trademark images, this paper proposes a trademark retrieval algorithm combining the image global features and local features. The global features capture the image gross contour. It can be used to rapidly build candidate image database. The local features can be used to more accurately match with candidate image. This paper extracts Fourier Descriptors（FDs） of the retrieved image and sorts them according to similarity. Candidate images are formed. The query image accurately match with candidate images using the SIFT features. Experimental results show that this method not only keeps SIFT features the perfect descriptive ability, but also reduces the computation complexity and has higher precision.

Key words: Content-Based Image Retrieval（CBIR）, trademark, Fourier descriptors, Scale Invariant Feature Transform（SIFT）

摘要： 利用商标图像的形状特征，提出了一种融合图像全局特征和局部特征的商标检索算法。其中全局特征反映了图像的整体信息，这些信息可用来较快地建立候选图像库，而局部特征则可以更准确地与候选图像进行匹配。提取图像的傅里叶描述子进行初步检索，按相似度排序，在此结果集的基础上对候选图像通过提取SIFT特征进行精确匹配。实验结果表明，该方法既保持了SIFT特征的良好描述能力，又减少了精确匹配需要的计算次数，降低了复杂度。

关键词: 基于内容的图像检索, 商标, 傅里叶描述子, 尺度不变特征转换（SIFT）