基于联合不变矩形状特征的商标图像检索

Trademark image retrieval based on united invariant moment shape feature

针对商标图像结构越来越复杂的特点,提出了一种基于联合不变矩形状特征的商标图像检索方法。首先利用最佳阈值分割和形态学腐蚀运算的轮廓提取进行精确边缘检测;然后计算16个联合不变矩组成图像的特征向量;最后对特征向量高斯归一化,用欧氏距离度量图像间的相似性,构建了一个基于形状特征的自动图像检索系统。实验结果表明,该方法具有一定的可行性和有效性,所构建的系统具有实用意义。

A method of trademark retrieval based on united invariant moment shape feature is proposed, considering the complexities of the trademark image structures. Firstly, the image edges are detected accurately with the both the optimal threshold segmentation and contour extraction of morphological corrosion computing. Then 16 united moment invariants are calculated to constitute the feature vectors of image. Finally, similarities among trademark images are measured via the Euclidean distance and normalized moment vectors to build an image retrieval system based on shape features. Experimental results demonstrate that the method is practical and effective.