基于局部和全局特征融合的图像检索

Image retrieval based on combining local and global features

为了有效地组织、管理、浏览、检索图像数据库,提出了一种综合全局统计特征和局部二值位图特征的图像检索算法。分别计算图像R、G、B三通道的均值和方差,获取了图像的全局统计特征。然后,根据块截断编码思想,将图像划分成4×4的图像子块,同样计算其均值。若块均值大于图像全局均值,则该块设为"1",否则,设为"0",由此,得到图像的二值位图特征。最后,对归一化的特征进行有机融合并采用最佳相似匹配函数进行检索。实验结果表明:综合两种特征的效果比使用单一特征的效果好;和同类算法相比,该算法鲁棒性好,前100幅图像的平均检索准确率达到63%,相对本文提到的另外两种算法都提高了4%以上。

In order to organize, manage, browse and retrieve image database effectively, a novel image retrieval algorithm combining global statistical feature and local binary bitmap feature was proposed. The mean value and standard deviation of every image were calculated to obtain the global statistical feature. Then, according to the idea of block truncation code, every image was divided into 4×4 image blocks without overlapping, and mean value of every image block was calculated also . If mean value was larger than that of the whole image, the image block was set as ＂1＂, otherwise, the image block as ＂0＂,so the image bitmap feature was obtained . Finally, the normalized features were integrated and retrieved by the best similar matching function. Experimental results indicate that the retrieval performance using combined global and local features is prior to that using single feature; compared with other similar schemes, the algorithm is robust, and the mean retrieval precision reaches 63% in the first 100 images, which is 4% higher than that of other two schemes mentioned in the paper.