摘要
分析了传统的基于小波边界点图像检索方法的缺点,提出了一种基于图像特征点环形分布概率的图像检索算法.该方法首先利用粗分辨率下绝对值大的小波系数对应于原始信号中全局变化大的区域这一特性,反向查找细分辨率下能够反映这种全局变化的小波系数,直到找到原始信号中相应的特征点;然后,将图像沿半径方向等分成若干同心圆环,计算图像特征点在这些圆环中的分布概率,以此作为图像的特征向量;最后,通过带权的街区距离计算图像间的相似度.实验表明本文提出的检索算法及图像相似度计算方法不仅检索精度高,还具有较优的旋转、尺度及视觉角度不变性.
A novel CBIR algorithm based on ring-shaped distribution of salient points are presented in this paper. First, the salient points were detected by wavelet transform, in whieh those wavelet coefficients can be calculated in different multi-resolutions. Then the location and some low level features such as color of salient points are extracted to describe each image patch. Secondly, the image is divided into a number of concentric rings along the radius direction, and the image feature vectors are computed according to the probability of salient points in these rings. Finally, similarity between images is calculated by the block-distance with weight. Experiments show that the retrieval algorithm and image similarity calculation method proposed in the paper is not only retrieval of high precision, but also excellent havariance feature in rotation, scale and visual angle of view.
出处
《计算机系统应用》
2013年第6期81-85,共5页
Computer Systems & Applications
关键词
小波变换
特征点
概率
距离
权
wavelet transform
salient point
probability
distance
weight
