一种基于改进Hu矩的图像检索方法

提出了一个在形状轮廓的基础上,进行改进的Hu矩的形状特征提取与检索方法。首先,对图像进行预处理,利用多阈值颜色分割法提取出图像的前景对象,再采用轮廓跟踪法得到该对象的形状轮廓特征;然后,计算7个不变矩,再转化为10个改进的Hu矩,组成图像的特征向量;最后,用欧氏距离度量图像间的相似度。实验结果表明,该方法具有较好的检索效果。

A Method of Image Retrieval Based on Improved Hu Moments

A new method for CBIR based on shape relative moments feature has been presented. Firstly, the image has been pre -processed, the foreground object has been extracted using multi- threshold image segmentation. Then the seven invariant moments have been calculated and translated to ten improve Hu moments, which constructors characterize vectors of image ; Finally, the similarity between two images＇ normalized moment vectors has been measured taking use of Euclidean distance. Experimental results showed that the method has a very good retrival effect.