摘 要: 針對(duì)航空序列擺掃圖像中心傾斜投影視角變化大的問題,提出一種基于中心投影變換與SIFT相結(jié)合的圖像拼接方法。該方法首先根據(jù)擺掃機(jī)構(gòu)提供的參數(shù),結(jié)合中心投影構(gòu)像方程計(jì)算出圖像的投影變換矩陣,用于對(duì)序列擺掃圖像進(jìn)行校正;運(yùn)用SIFT算法提取圖像重疊區(qū)域的特征點(diǎn),并采用歐氏距離進(jìn)行特征點(diǎn)匹配;然后利用RANSAC算法剔除錯(cuò)誤匹配點(diǎn),并計(jì)算出變換矩陣對(duì)圖像進(jìn)行配準(zhǔn);最后采用漸入漸出的圖像融合方法得到無(wú)縫拼接的圖像。通過與傳統(tǒng)圖像拼接方法進(jìn)行對(duì)比,實(shí)驗(yàn)證明文中提出的方法較大地提高了配準(zhǔn)的精度以及拼接效果。
關(guān)鍵詞: 投影變換; SIFT; 圖像配準(zhǔn); 融合方法; 圖像拼接
中圖分類號(hào): TN911.73?34; TP391 文獻(xiàn)標(biāo)識(shí)碼: A 文章編號(hào): 1004?373X(2015)09?0059?06
Abstract: To solve the problem, that the center oblique projection has great variation in angle of view, an image mosaic method base on center projection transformation and SIFT is presented for scanning image sequence. First of all, according to the parameters provided by the scanning mechanism and the center projection imaging equation, the projection transformation matrix which is applied to correcting the sequence scanning image is calculated, the SIFT algorithm is utilized to extract the feature points of the image overlap area, which are matched by using Euclidean distance, and then the RANSAC algorithm is used to eliminate the error matching point and calculate transformation matrix for image match. Finally, the seamless mosaic image is obtained by the image fusion method. Compared with the traditional fusion method, the results show that the accuracy of image match and image mosaic effect are improved by the proposed image mosaic method.
Keywords: projection transformation; SIFT; image match; fusion method; image mosaic
0 引 言
隨著無(wú)人機(jī)航拍技術(shù)的發(fā)展,無(wú)人機(jī)航拍技術(shù)越來(lái)越多地應(yīng)用到軍事、測(cè)繪、環(huán)境監(jiān)測(cè)等領(lǐng)域。而航空相機(jī)是人們獲取地面信息的重要手段,為了擴(kuò)大相機(jī)視場(chǎng),可以采用多個(gè)CCD相組合,但是這樣大大增加了相機(jī)的體積、質(zhì)量、成本和復(fù)雜度[1?2]。所以可以將框幅式相機(jī)安裝于轉(zhuǎn)臺(tái)上,通過轉(zhuǎn)臺(tái)的運(yùn)動(dòng)實(shí)現(xiàn)擺掃成像,這時(shí)就需要圖像拼接生成大視場(chǎng)的圖像。
圖像拼接包括圖像預(yù)處理、圖像配準(zhǔn)、圖像融合,其中圖像配準(zhǔn)是關(guān)鍵。圖像配準(zhǔn)一般分為基于像素與基于特征兩種方法,但前者易受光照變化影響且計(jì)算量較大。而David G.low提出的基于特征的SIFT方法因?yàn)榫哂衅揭?,旋轉(zhuǎn),縮放以及光照不變性等優(yōu)點(diǎn)[3?4],得到了廣泛的應(yīng)用,但大視角變換會(huì)對(duì)拼接效果造成一定的影響[5?6]。另一種圖像拼接的方法是在內(nèi)外方位元素已知的情況下,利用中心投影共線方程精確校正傾斜相片,直接完成拼接,但是該方法需要精確的主動(dòng)姿態(tài)測(cè)量設(shè)備,并且需要標(biāo)定。針對(duì)以上問題本文提出一種基于中心投影變換與SIFT結(jié)合的圖像拼接方法,該方法通過主動(dòng)姿態(tài)測(cè)量設(shè)備提供的部分參數(shù),結(jié)合中心投影構(gòu)像方程計(jì)算出圖像的投影變換矩陣,對(duì)序列圖像進(jìn)行粗校正,再利用SIFT算法進(jìn)行特征提取完成圖像拼接,不僅克服了以上圖像拼接的問題,而且較大地提高了圖像配準(zhǔn)精度與拼接效果。
1 基于投影變換序列圖像幾何校正
通過擺掃生成的序列圖像有較大的視角變化,根據(jù)擺掃成像原理越靠近邊緣的圖像其擺掃角度越大,畸變也越大。本文首先通過中心投影構(gòu)像方程推導(dǎo)出投影變換矩陣,并利用矩陣變換對(duì)序列圖像進(jìn)行粗校正用于后期圖像拼接。
3 圖像拼接評(píng)價(jià)函數(shù)
5 結(jié) 論
由于序列擺掃圖像有較大的視角變化,直接對(duì)序列圖像進(jìn)行拼接配準(zhǔn)精度較低,甚至?xí)斐善唇渝e(cuò)誤。本文提出的對(duì)序列擺掃圖像先利用投影變換矩陣進(jìn)行幾何校正再拼接的方法,有效地解決了視角差太大的問題,提高了配準(zhǔn)精度以及圖像拼接的效果,當(dāng)擺掃圖像的擺掃角度增大時(shí),本文提出的拼接方法更有優(yōu)勢(shì)。
參考文獻(xiàn)
[1] PETRIE G, WALKER A S. Airborne digital imaging technology: a new overview [J]. The Photogrammetric Record,2007,22(119): 203??225.
[2] WANG Hui, WU Yun?dong, LIANG Hong. Research status of digital aerial camera system [J]. Journal of Zhengzhou Institute of Surveying and Mapping, 2007, 24(3):171?174.
[3] LOWE D G,Object recognition from local scale?invariant features [C]. Proceedings of the 7th IEEE International Confe?rence on Computer Vision. Kerkyra, Greece: IEEE, 1999, 2: 1150?1157.
[4] LOWE D G. Distinctive image features from scale?invariant keypoints [J]. International Journal of Computer Vision, 2004, 60(2): 91?110.
[5] MIKOLAJCZYK K, SCHMID C. A performance evaluation of local descriptors [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2005, 27(10):1615?1630.
[6] HUANG Wei, WEI Ying?mei, XIE Yu?xiang, et al. Survey of local invariant feature description [C]// Chinese Automation Congress (CAC), 2013. Changsha: IEEE, 2013: 353?358.
[7] 朱述龍,珠寶山,王紅衛(wèi).遙感圖像處理與應(yīng)用[M].北京:科學(xué)出版社,2006.
[8] 楊化超.圖像局部不變特征及其匹配問題研究及應(yīng)用[M].北京:測(cè)繪出版社,2013.
[9] CAPEL D P, Image mosaicing and super?resolution [D]. Oxford: Robotics Research Group Department of Engineering Science University of Oxford, 2001.
[10] KOENDERINK J. The structure of image [J]. Biological Cybernetics, 1984, 50(5): 363?370.
[11] FISHIER M A, BOLLES R C. Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography [J]. Comm. Assoc Comp Mach, 1981, 24(6):381?395.
[12] LI Bin, MING De?lie, YAN Wen?wen, et al. Image matching based on two?column histogram hashing and improved RANSAC [J]. IEEE Geoscience and Remote Sensing Letters, 2014, 11(8): 1433?1437.
[13] 程光權(quán),成禮智,趙俠.基于幾何特征的圖像處理與質(zhì)量評(píng)價(jià)[M].北京:國(guó)防工業(yè)出版社,2013.
[14] WANG Hong?nan, ZHONG Wen, WANG Jing, et al. Research of measurement for digital image definition [J]. Journal of Image and Graphics, 2004, 9(7): 828?832.