Precision Motion Detection Algorithm using Neural Networks

使用神经网络的精确运动检测算法

• 批准号: 13650411
• 负责人: YAMADA Yoshio
• 金额: $ 2.62万
• 依托单位: EHIME UNIVERSITY
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (C)
• 财政年份: 2001
• 资助国家: 日本
• 起止时间: 2001 至 2002
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-13650411/
• 关键词: Template Matching; Motion Estimation; Neural Network; Cross Correlation; Peak Detection; White Gaussian Noise; 相互相関; 相互相関関数; 遺伝的アルゴリズム

Template matching is one of the essential image processing techiniques used in the areas such as pattern recoginition and image analysis, etc. Continuous template matching method, which is referred to as the conventional method hereinafter, was already proposed by the investigators of this project in order to enable precision detection of small displacements between two digital images. In the proposed method, a steepest descent method based successive approximation algorithm is used to detect the peak position of bandlimited interpolated continuous function. And this leads to the problem in computaional complexity and fluctuation of computational time. In order to solve the problem, a novel neural networks based template matching method is proposed in this research project. Accuracy and computational complexity of the proposed method are evaluated by computer simulations and compared with those of the conventional method. As a noise model for input images, as the first order approximation, the additive white Gaussian noise is assumed. It is found that, by using the proposed method, the computational complexity of peak position detection is reduced to 1/13 as compared with that of the conventional method without loss of the accuracy of displacement detection.

模板匹配是一种重要的图像处理技术，广泛应用于模式识别和图像分析等领域。为了精确检测两幅数字图像之间的小位移，本项目研究者已经提出了连续模板匹配方法，以下简称常规方法。该方法采用基于最陡下降法的逐次逼近算法来检测限带插值连续函数的峰值位置。这就导致了计算复杂度和计算时间波动的问题。为了解决这一问题，本课题提出了一种基于神经网络的模板匹配方法。通过计算机仿真对该方法的精度和计算复杂度进行了评价，并与传统方法进行了比较。作为输入图像的噪声模型，假设加性高斯白噪声作为一阶近似。结果表明，在不影响位移检测精度的前提下，该方法将峰值位置检测的计算复杂度降低到传统方法的1/13。

项目成果

Yoshio YAMADA and Shinji TSUZUKI: "Continuous Template Matching Method Using Neural Networks"J.of ITE. Vol.57, No.3. 399-402 (2003)

Yoshio YAMADA 和 Shinji TSUZUKI：“使用神经网络的连续模板匹配方法”J.of ITE。

山田芳郎, 都築伸二: "ニューラルネットワークを用いた連続的テンプレートマッチング手法"映像情報メデイア学会誌. 53. 399-402 (2003)

Yoshiro Yamada、Shinji Tsuzuki：“使用神经网络的连续模板匹配方法”视觉信息和媒体工程师学会杂志 53. 399-402 (2003)。

山田芳郎, 都築伸二: "ニューラルネットワークを用いた連続的テンプレートマッチング手法"映像情報メディア学会誌. 53. 399-402 (2003)

Yoshiro Yamada、Shinji Tsuzuki：“使用神经网络的连续模板匹配方法”图像信息和电视工程师学会杂志 53. 399-402 (2003)。