摘要
本文提出了一种基于非张量积小波分解的高效图像匹配方法。首先利用一种非张量积小波将图像模板和被搜寻区域图像分解成粗尺度图像数据;通过自适应阈值选取算法最终完成匹配,并可将搜寻到的目标图像定位到原图上。然后,在粗尺度被搜寻图像上找到的候选目标区域再在细尺度图像上做进一步匹配,直到满足一定的精度要求。文末给出了实验。
An efficient approach for image matching in large size 2-D images is proposed. Low-frequency band images produced by a kind of nonseparable compactly supported orthonormal continuous wavelet decomposing is used to represent the feature of the target. We then regard both decomposed template image and searching region image as vectors and define a similarity between them. We also design a self-adaptive threshold selecting algorithm to complete the image matching in small-size decomposed images, and the result can be located in the original image. The candidate targets found in large-scale images can be matched in small-scale for higher precision. An experiment of finding a target building from a natural scene image with high random noise (30%) and brightness perturbation is shown to demonstrate the efficiency of our approach. We also show advantages of nonseparable wavelets by comparing it to Daubechies wavelets in an application of integrated circuit layout inspection.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
1999年第1期67-73,共7页
Pattern Recognition and Artificial Intelligence
基金
国家科委重点科技项目"亚微米"
深亚微米集成电路自动化分析技术研究课题之一
关键词
非张量积
小波
图像匹配
图像识别
Multi-Scale, Template, Match, Nonseparable, Wavelet, Similarity
