摘 要:SIFT算子在實際應用中,由于地面圖像本身特征不明顯且提取出的特征點多、亂以及灰度變化不明顯等特點的影響,從而導致特征點誤匹配。為此提出一種改進的SIFT圖像特征匹配算法。該算法是在SIFT特征匹配的基礎上,利用多目標優化算法,建立相關匹配模板,利用給定同一場景的兩幅圖像, 尋找同一場景點投影到圖像中的模板之間的相關性建立數學模型即目標函數,根據同一幅圖像中模板間的距離建立邊界約束條件,從而剔除一些誤匹配點。實驗表明,該算法可以有效地提高圖像匹配精度。
關鍵詞:SIFT; 多目標優化; 特征匹配; 誤匹配點
中圖分類號:TN911.73 文獻標識碼:A
文章編號:1004-373X(2010)12-0099-04
SIFT Feature Matching Algorithm Based on Multi-objective Optimization
YAN Zhi, WANG Li-ming,CHEN Ping
(National Key Laboratory for Electronic Measurement Technology, North University of China, Taiyuan 030051, China)
Abstract:SIFT (Scale Invariant Feature Transform) descriptor is commonly used in image matching due to the invariance of scale, rotation and illumination. While in practical applications, incorrect matching of feature points was happened because the ground image′s features were not obvious and the feature points extracted by SIFT were more, chaos as well as gray-scale did not change significantly. In order to solve the problem, an improved SIFT image feature point matching method is proposed. The method based on SIFT feature matching algorithm and used multi-objective optimization algorithm to establish correlation matching template, the two images of same scene was given to find an image projected onto a template to establish the correlation between the mathematical model of the objective function. According to distance between the template in the same image to establish the boundary constraint conditions, thus eliminate some 1 matching points. The experiments show that the algorithm can effectively improve its accuracy.
Keywords:SIFT;multi-objective optimization;feature matching; incorrect matching point
0 引 言
圖像特征匹配是計算機視覺和模式識別等領域研究的基本問題以及物體識別、跟蹤等應用的重要基礎,廣泛應用于攝影測量與遙感、資源分析、三維重建、目標識別、圖像拼接、圖像檢索等眾多領域, 一直是研究者關注的焦點, 在針對地面圖像的特征匹配方面得到了深入研究,以及對運動目標軌跡提取具有較高的研究意義和應用價值。但是由于它受到天氣、陽光、遮擋等外界因素的嚴重影響, 并且存在因不同的成像時間、角度、距離等因素而導致的圖像平移、旋轉、縮放等問題, 這都給目標特征檢測及特征匹配的工作帶來了很大的難度[1] 。針對這些不足,尤其是在尺度方面的不足,David G.Lowe在2004年總結了現有的基于不變量技術的特征檢測方法,提出了一種基于尺度空間的特征匹配算法——尺度不變特征變換(scale invariant feature transform,SIFT)算法[2-3] 。……