摘要： 提出了一种基于小波和相对矩的形状特征提取与检索方法。首先,对亮度图像进行小波多尺度边缘检测,得到多尺度边界图像;然后,计算每一尺度的7个不变矩,再转化为10个相对矩,所有尺度上的相对矩组成图像的特征向量;最后,对特征向量进行高斯归一化,用欧氏距离度量图像间的相似度,构建了一个基于形状特征的图像检索系统。实验结果表明,该方法具有明显的优越性和通用性。

关键词: 基于形状的图像检索, 多尺度边缘检测, 不变矩, 相对矩

Abstract: This paper presents a shape feature extraction and image retrieval method based on wavelet and relative moments.Firstly, it processes the luminance images with wavelet edge detection to get multi-scale edge images, then calculates the seven invariant moments of every scale, and translates to ten relative moments, which a group of multi-scale relative moments vectors in feature space characterize each image. Finally, the Euclidean distance between two images' normalized moment vectors gives the measure of similarity, and constructs a shape-based image retrieval system. Experimental results verify the superiority and currency of this method.

中图分类号: 

• TP391.41