目的 针对单一低层特征在语义属性中的信息易丢失，导致其对图像描述能力不强，使其检索精度不佳的问题，结合颜色矩（CM）、角径向变换描述符（ART）和边缘直方图（EH）等3种特征，定义一种双级特征提取与度量的图像检索方案。方法 首先，将图像转换为HSV色彩空间，并将其分割为若干个非重叠子图像，通通过计算每个子图像的均值、标准差和偏斜度来表征CM；再利用Euclidean距离，对查询图像和数据库图像的CM进行提取与度量，将输出的检索结果标记为一个图像集。随后，提取查询图像与图像集中每个目标的ART和EH特征；利用Euclidean距离分别度量查询图像与图像集中目标的ART与EH的相似性；最后，对ART与EH的加权组合，输出相似性最高的检索图像。结果 实验表明，与当前常见的检索算法比较，文中算法具有更高的检索精度，表现出更优异的Precision-Recall曲线。结论 所提算法具有良好的检索准确度，在信息处理、包装商标等领域具有一定的参考价值。

Aiming at the problem of information loss in semantic attribute of single low-level feature, the ability of autonomous learning is weak, which leads to poor image expression, three characteristics of Color Moment (CM), Angular Radial Transform (ART) and Edge Histogram (EH) were combined, an image retrieval scheme based on two stage feature extraction and metric was defined. Firstly, the image was transformed into HSV color space and divided into several non-overlapping sub-images. The mean, standard deviation and skewness of each sub image were calculated as CM representation. Secondly, Euclidean distance was used to extract and measure the CM of query image and database image, and output of query was marked as an image set. The ART and EH features of the query image and the image set obtained in the K results were extracted, and then the similarity between the ART and EH of the query image and the first stage image was respectively measured with Euclidean distance. By weighted combination of ART and EH, retrieval images with the highest similarity were output. Experiments showed that: compared with the current retrieval algorithms, the proposed algorithm had excellent precision of retrieval recall and superior Precision-Recall curve. This proposed algorithm has good retrieval accuracy, which has certain reference value in the fields of information processing and packaging trademark.