中文摘要：

目的解决图像形状信息复杂，特征描述不够准确所导致的检索精度问题。方法提出了一种基于Canny边缘提取的融合Hu不变矩及其推广的5个新不变矩作为形状的可视特征描述的方法，并通过归一化的特征向量的加权相似距离在图像形状检索过程中进行匹配。结果实验结果表明：该方法能够充分利用图像的形状信息，比单一提取Hu不变矩特征具有更好的检索性能。结论新推广的5个不变矩与原有的7个不变矩结合之后能够更为完整地描述图像的形状信息，并且能够提高检索性能。

英文摘要：

Aim To solve the problem of retrieval accuracy due to less precision description of features for shape complicated images. Methods A new method is presented, which integrates 7 Hu moment invariants based on the edge detection of Canny operator with 5 new of their promotion as the image visual shape features＇description. Its application in content-based image retrieval （CBIR） is also introduced. After calculating the weighed similarity dis- tance between the query image and each image in the image database through the normalized feature vectors, matched images can be found. Results Images＇shape information is used more sufficiently than singly using Hu moment invariants, and better retrieval performance is got. Conclusion The image is described more perfectly by integrating 5 new generalized moment invariants with 7 original Hu moment invariants, and the retrival performance is also improved.