基于区域伪Zernike矩的商标图像检索

Trademark Image Retrieval Using Region Pseudo-Zernike Moments

研究商标图像检索问题,针对在二值商标图像检索中要求准确、高效,因此确定出目标像素点的区域分布形式,解决商标形状特征,对平移、旋转、比例以及起始点等变化下特征不变性问题,提出了一种基于区域伪Zernike矩的提取图像形状特征的算法。算法根据图像目标区域的最小外接圆进行区域划分,提取区域的伪Zernike矩来描述图像形状特征进行商标检索。通过仿真实验证明,区域伪Zernike矩的提取图像形状特征的算法具有良好的平移,旋转及比例不变性,并用有较强的抗噪性。

In the binary-trade-mark retrieval,the regional distribution of the target pixel point is the important information in shape description,and for two similar binary-trade-mark images,the regional distribution of the target pixel point is also similar.However,the problems in the invariance property under translation,rotation,scale and the starting point has not been solved.In view of these problems,a new trademark image retrieval method based on region pseudo-Zernike moments is presented in this paper.The image object region is extracted according to the object pixels＇ circum-circle and then partitioned based on the concentric circles to produce a series of sub-images.The pseudo-Zernike moments of sub-images are computed,which are constructed into a feature vector for describing the shape of the trademark image.Experiment results show that this method has good invariant property under translation,rotation and scale,and is insensitive to noise.