摘要
基于区域的图像检索大都是利用各个区域的高频部分来直接代表图像的纹理特征,这种算法可能会导致诸如纹理平移等现象,增大图像的距离度量的误差,降低了检索的精度。本文提出了一种基于高阶微分的纹理特征选取算法,有效地解决了这类问题。另外,本文对Meanshift分割算法进行了改进,减小了由于区域吞噬而带来的颜色度量的误差。实验结果表明这种方法是有效的。
RBIR uses the coefficients of high frequencies of every region to represent the texture feature of image, which causes the problem of motion of the texture window, increasing the distortion of the distance metric, reducing the precise of the retrieval. We propose a high-order differential coefficient method, it solves this problem effectively. Furthermore, this paper improves the Meanshift algorithm, reduces the distortion of the licking up of regions. It is effective in experiment.
出处
《微计算机信息》
北大核心
2007年第18期302-303,226,共3页
Control & Automation
基金
广东省自然科学基金(No.05006349)
