不同時期影像中地物變遷情形偵測為GIS資料庫更新時之重要來源，因此許多研究針對此方向進行影像中之變遷偵測。於航空影像中變遷偵測所使用的方法大致可分為依光譜特性之變遷及幾何特性之變遷。前者較多而後者較少，幾何變遷的研究中主要以兩張影像萃取出之線性特徵直接進行匹配為主。由於不同時期影像拍攝時，受各種因素的影響使得兩張影像灰值並不一致，即使使用相同的特徵偵測模式於影像中相同地物上可能萃取出不同的線特徵資訊，若以該線特徵資訊直接進行匹配來判定是否變遷時將產生錯誤的結果。
　　為了避免造成上述錯誤的結果，本研究在使用影像中線特徵進行變遷偵測時，提出修正的方案。本研究不直接匹配二不同時期影像所萃取出的線段，而是改為利用一期影像中萃取出之線段鏈套合於另一期影像對應位置上尋找該位置上是否有線段存在可能性，如果沒有則判斷是已變遷。但當影像外方位不是很精確或地物受高差移位影響時使得套合位置本身不精確，以致必須擴大尋找範圍，有可能使得在可能套合範圍內的線段並非單一，本研究試著以機率式鬆弛法計算線段鏈與所有可能被套合線段間之匹配機率來區分它們。
　　為了測試本文所提方法之可行性，我們選取不同時期航空影像進行實驗。由實驗結果中可看出，以梯度萃取局部線段方法可成功偵測出與線段鏈套合位置近似之所有線段。加入機率式鬆弛法後計算出線段鏈與局部線段最高匹配機率可用於變遷偵測之參考。

　　Changes of contents in aerial images from different epochs are important clues for updating GIS database. Many researches have dealt with this topic in the past. Approaches of change detection can generally be divided into two classes, namely the spectral approach and the geometric feature approach. The latter one finds changes in images that were taken at different epochs, by comparing extracted line features. This comparison is achieved usually through matching of line features from one epoch to another directly. But this method is not very effective. Because images taken at different epochs usually have very different gray value behavior even when their contents are exactly the same. That means, even when the same feature detector with exactly the same thresholds and control parameters were used to extract line features of images having no changes, different line could still be detected. This will of course give false alarms for changes.
　　It is under this consideration that this research uses a different philosophy to detect line feature changes. Our method is based on evidence finding. Instead of trying to match the extracted line features directly, our method tries to prove (or disapprove) the existence of a line feature at the corresponding location where, in the other image, a line feature has been extracted. It is well known that, based on the principle of edge detection, if at any places there should be a line feature, there must be a local peak in the gradient value, no matter how strong or how weak this peak might be. Therefore finding evidences is equivalent to looking for the existence of local gradient peaks. When the exterior orientations of both images are known, the corresponding locations in both images are easily to be found. But if the exterior orientations of the two images are not known very well, the corresponding location of the line feature of one image in the other image could not be known very accurate. In this case the search area for local gradient peak must be enlarged. It is very probable that more than one line feature could exist in the search area. In order to cope with this problem, we use the probabilistic relaxation matching to find the best match between two images.
　　In order to test the feasibility of this thought, we have made some experiments using two aerial images taken at different epochs. Result of the experiments show that our evidence finding approach is feasible and the use of the probabilistic relaxation does improve the chances of finding the correct match.

參考文獻
邱式鴻，王蜀嘉。Förstner特徵物萃取法精確性探討，第十五屆測量學術及應用研討會，p667-676，1996。
邱式鴻，從都市區立體航照影像中萃取屋頂面的實用策略，國立成功大學測量工程研究所博士論文，2001。
張祖勛，張劍清，數字攝影測量學，武漢測繪科技大學出版社，1997。
陳順宇，鄭碧娥，統計學，華泰書局，1998。
陳建文，彩色影像邊緣線萃取之研究，國立成功大學測量工程研究所碩士論文，2002。
Agrouris, P., Beard, K., Mountrakis, G.. and Stefanidis, A., Capturing and Modeling Geographic Object Change: A SpatioTemporal Gazetteer Framework. Photogrammetric Engineering and Remote Sensing, 66: 1241-1250，2000.
Canny, J., A Computational Approach to Edge Detection, IEEE Tran. Pattern Analysis and Machine Intelligence, 8(6): 679-698，1986.
Carroll, R. W. , Detection Buildings through Imagery and Automatic Feature Processing, To be presented at the Integrating GIS & CAMA 2003 Conference，2003.
Christmas, W. J., Kittler, J. and Petrou, M. , Structural Matching in Computer Vision Using Probabilistic Relaxation, IEEE Transaction on Pattern Analysis Machine Intelligence, 17(8): 749-764，1995.
Curz, J. M., Pajares, G. and Aranda, J. , A Neural Network Approach to the Stereovision Correspondence Problem by Unsupervised Learning, Neural Networks, 8: 805-813，1995.
Douglaus, D. H. and Peucker T. K. , Algorithms for Reduction of the Number of Point Required to Represent a Digitized Line or its Caricature, Canadian Cartographer, 10: 110-112，1973.
Förstner, W. , A Framework for Low Level Feature Extraction, Ekiundh, J.-O. (Ed), Computer Vision, ECCV’94, Lecture Notes in Computer Science, 801, Springer-Verlag, Berlin, 11: 383-394, 1994.
Gonzalez, R. C. and Wood, R. E. , Digital Image Processing, Addison-Wiley, New York，1992.
Huertas, A. and Nevatia, R. , Detecting Changes in Aerial Views of Man-Made Structures, Proceedings, International Conference on Computer Vision. pp.73-80，1998.
Hoffmann, A., van der Vegt, J. W. and Lehmann, F. , Towards Automated Map Updating: is it feasible with new digital data acquisition and processing techniques, International Archives of Photogrammetry and Remote Sensing XXXIII(B2): 295-302，2000.
Jung, F. , Detection Building Changes from Multitemporal Aerial Stereopairs, ISPRS Journal of Photogrammetry & Remote Sensing, 58: 187-201，2004.
Knudsen, T. and Olsen, B.P. , Detection of Buildings in Colour and Colour-Infrared Aerial Photos for Semi-Automated Revision of Topographic Databases, ISPRS-Commission III Symposium，2002.
Nasrabadi, N. M. , A Stereo Vision Technique Using Curve-Segments and Relaxation Matching, IEEE Transaction on Pattern Analysis Machine Intelligence, 14(5): 566-572，1992.
Pajares, G., Cruz, J.M. and Lopez-Orozco, J.A. , Relaxation Labeling in Stereo Image Matching, The Journal of the Pattern Recognition Society, 33(1): 53-68，2000.
Peuquet, D. J. , An Examination of Techniques for Reformatting Digital Cartographic Data / Part 1: The Raster-to-Vector Process, Cartographica, 18(1): 34-48，1981.
Rosenfeld, A., Hummel, R. A. and Zucker, S. W. , Scene labeling by relaxation operations, IEEE Trans. Systems Man Cybernet, 6(6): 420-433，1976.
Row, N. C. and Grewe, L. L. , Change Detection for Linear Features in Aerial Photographic Using Edge-Finding, IEEE Transactions on Geoscience and Remote Sensing, 39(7): 1608-1612，2001.
van Asperen, P. , Digital Updates at the Dutch Topographic Service, International Archives of Photogrammetry and Remote Sensing, 31(B4): 891-900，1996.
Walter, V. and Fritsch, D. , Automatic Verification of GIS Data Using High Resolution Multispectral Data, International Archives of Photogrammetry and Remote Sensing XXXII(3): 74-79，1998.
Walter, V. and Fritsch, D. , Automated Revision of GIS Databases, Proceeding of the Eight ACM Symposium on Advances in Geographic Information Systems, pp. 129-134，2000.