Automatic Target Detection from Battle Scene Images Using 2D Principle Component Analysis

CNU IR > College of Smart Living > Dept. of Information Management > Periodical Articles > Item 310902800/26944

Please use this identifier to cite or link to this item: https://ir.cnu.edu.tw/handle/310902800/26944

Title:	Automatic Target Detection from Battle Scene Images Using 2D Principle Component Analysis 運用二維主成分分析法於戰場影像的自動目標偵測
Authors:	Lee, Jen-Chun Ping, S.Huang Yang, Sheng-Kai Chiang, Chung-Shi
Contributors:	資訊管理系
Keywords:	Automatic target detection (ATD) Eigenspace separation transform (EST) Two-dimensional principle component analysis (2DPCA) 自動目標偵測特徵空間分離轉換二維主成分分析法
Date:	2011-11
Issue Date:	2013-09-28 16:25:12 (UTC+8)
Abstract:	Automatic target detection (ATD) is of crucial interest to the defense community. This paper describes a novel ATD approach from battle scene images by using two-dimensional principle component analysis (2DPCA). The proposed algorithm efficiently utilizes the target intensity feature (inner image vector) and surrounding background (outer image vector) for detection purposes. Firstly, these image vectors are preprocessed by high-pass filters to get image gradient vectors. Secondly, we employ 2DPCA and eigenspace separation transform (EST) to obtain two linear functions, respectively. Finally, the difference of the two linear functions is used to extract the target from the background. Experimental results by applying this approach at real-life short-range and medium-range battle scene images sequences demonstrate that the method can greatly reduce the computation complexity and improve the rate of target detection. 自動目標識別系統為目前在軍事防衛系統中的重要課題。本論文提出一個新穎的自動影像目標識別演算法，主要係運用二維主成分分析法(2-dimentional principal components analysis, 2DPCA)來針對戰場上的目標影像進行自動偵測。首先，運用高通濾波器將場景影像轉換為梯度影像。然後運用二維主成分分析法與特徵空間分離轉換法 (Eigenspace Separation Transform, EST)分別找出兩個線性函數。最後，我們利用這兩個線性函數來區別出影像區塊為目標或背景。在實驗中，我們分別使用了一系列之戰場短距離及中距離影像來進行演算法驗證，結果證明所提出之方法，無論在速度及準確度上，均較原有方法有較佳之改良結果。
Relation:	中正嶺學報 40(2), pp.79-90
Appears in Collections:	[Dept. of Information Management] Periodical Articles

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	2761	View/Open

Loading...