Development of Precision Template Matching Method and Its Application to Motion Detection of Image Sequences

本项目开发了一种新的模板匹配方法，旨在高分辨率检测两幅数字图像之间的位移，其中高分辨率检测意味着检测与采样或像素周期相比非常小的位移的能力。本文提出并讨论了带限FIB连续模板匹配存储器的几个关键方程，和双线性插值模型。讨论了采用高通滤波器（HPF）进行预滤波对保证连续模板匹配精度的必要性。该项目的结果总结如下。(1)通过计算机仿真对三种插值模型进行了比较，发现限带插值模型的精度最高。(2)本文将三种插值模型下的连续模板匹配方法转化为离散模板匹配方法，并在此基础上提出了一种仅使用离散样本的通用高效模板监视算法。(3)实验结果表明，该方法结合带限插值模型，可获得1/10 ~ 1/100像素的分辨率。(4)计算效率，两步匹配算法的开发。(5)提出了应用于数据嵌入方法的可能性。

This project has developed a new template matching method aiming at a high-resolution detection of displacements between two digital images, wherein the the high-resolution detection means the ability to detect very small displacement as compared with sampling or pixel period. Several key equaltions are developed and discussed for continuos template matching memods with band-limited, FIB., and hi-linear interpolation models. It is also discussed that the pre-filtering using High-Pass Filter (HPF) is essential for ensuring the precision of the continuous template matching. Results of the project are summarized s follows. (1) Investigating the three interpolation models by computer Simulation it is found that bandlimited interpolation model gives best precision. (2) Continuous temuplate mathing methos with three interpolation models are formulated and modified into discrete-type, and then comutiotionally efficient template watching algorithms only using discrete Samples. (3) Using test image sequences, 1/10 - 1/100-pixel resolutions are obtained by proposed method with band-limited interpolation model. (4) Computationally efficient, two-step matching algorithm are developed. (5) Possibility of application to data embedding method is presented.