一种基于目标区域的图像检索方法

为了弥补颜色直方图等全局特征在描述彩色图像空间信息上的不足,该文提出了一种基于目标区域的彩色图像检索方法。该方法首先利用一种基于颜色视觉一致性的图像分割方法提取出有意义的目标区域,然后分别对各个区域提取HSV彩色直方图和Hu不变矩作为目标区域的特征描述,最后提出了一种相应的计算相似度的方法,实现了图像之间的相似度度量。通过在中科院计算所的Mires图像数据库和ViViLab测试图像库上进行实验,该文提出的方法对于目标明确、背景不太复杂的图像可以达到较好的检索效果。

An Image Retrieval Method Based on Region

Recently,the global features like color histogram are used widely in content-based image retrieval systems(CBIR),these features can describe the global feature effectively,however,they can't describe the local spatial feature of image effectively.To overcome this shortcoming,this paper proposes a new algorithm based on region.First,this algorithm divides the image to multi-regions by a segment method based on perceptual color cluster;then,extract the color histogram of HSV color space and Hu moment in every regions;finally,this paper implements similarity measurement of two images by taking into consideration the similarity of every region synthetically.By the experimentation in the Mires image database,new algorithm can work effectively and efficiently.