| 摘要: | 摘要： 地形線形特徵萃取於地球科學的研究中佔有極重要的份量，過去對於線形之萃取由繪圖員，以手繪的方式於影像中描繪出線形。隨者影像的解析度日漸增加，影像之數量與日遽增，過去人工手繪的方法不只緩慢，而且存在有許多主觀判斷，不利後續統計研究。本研究之目的在於利用電腦自動判識，建立一個快速，客觀之方法，以期能夠協助大量處理影像，並且提供客觀之線形資料。 合成孔徑雷達（SAR）影像中存在許多特徵，包含線形與非線形之特徵，本研究利用小波轉換（wavelet transform）以及空間相交法(spatial correlator)將影像中之特徵萃取出來，接者利用霍氏轉換（Hough transform）將所萃取出來的特徵影像中屬於線形之部分連接起來。小波轉換的原理在於將影像中快速變化的部分視為特徵部分，變化較慢的部分則否。首先利用小波轉換產生一組多尺度影像，每一尺度影像均包括垂直、水平及對角等三張小波轉換之正交基底的高通系數影像，再藉由空間相交法累乘不同尺度影像， 將地圖上之的特徵部分辨識出來，使其以點的形式存在於空間之中。 霍氏轉換（Hough transform），作為線形萃取之工具，將特徵影像上之點轉換另一參數空間中之線，接者在將參數空間中多線之交集所得之點轉換回原始影像中之線段，如此可將於原來存在空間上之點轉換為線形，完成線形之萃取。 Abstract: The linear of the earth's surface occupies the extremely important part in the research of geosciences. In the past extraction linear of the image usually work done by hand. However, resolution of the image increases day by day, and the quantity of the image grow with each passing day. Extraction linear of the image work done by hand not only slow, but also have a lot of subjective judgement, and unfavorable follow-up study. In the study will use computer to automatically extraction linear of the image. Set up one fast, objective method; expect punishing the image in a large amount. And of far one objective linear date. Synthetic aperture radar（SAR） image have a lot of feature include non-linear and linear, in the study will use wavelet transform and spatial correlator get the feature points in the image, and use Hough transform to connect feature points of collinear in the feature image. The principle that wavelet transform is in regarding the part changed fast in the image as the feature part, change slower part to deny, In the study ,in fist , to produce multiscale images of one group to utilize wavelet transform. Every image of multiscale images of one group includes the high pass coefficient image of orthogonal in the basis that three, such as vertical, the level and opposite angle. And use spatial correlator to multiply different multiscale images. To get feature points in the feature image After making the feature image, and then, Hough transform as the pure tool that extraction linear. Changes the points on the feature space into the line of parameter space, after that change the point of many line to cross on parameter space into linear on feature space, And be able to change points into linear in feature space , finish extraction linear. |
