一种有效的多特征图像检索方法

An Effective Image Retrieval Method Based on Multi-Features

针对已有基于单一特征和综合特征的图像检索方法普遍存在的检索准确率较低、检索结果中的相关图像排序靠后的问题,提出一种有效的多特征图像检索方法.采用(16∶4∶4)量化方案对图像颜色进行量化,提取颜色特征,采用标记分水岭分割算法对图像进行分割,提取Jan Flusser不变矩作为形状特征,最后通过高斯归一化得到包含颜色和形状信息的多特征相似度.实验结果表明,该方法提高了检索准确率,使得检索结果中的相关图像排序靠前.

To improve the accuracy of single or integrated feature‐based image retrieval methods ,an effec‐tive image retrieval method based on multi‐features is proposed in this paper .A （16∶4∶4） quantization scheme is adopted to quantify image color and extract the color features of the image .The marked‐water‐shed algorithm is performed to achieve the segmentation of the image ,and then to extract Jan Flusser in‐variant moments .Finally ,Gaussian model is applied to normalize the different subcharacters distance .The final multi‐features similarity consists of the color similarity and the shape similarity .An experiment has demonstrated the efficiency of the proposed method .