期刊名称：Archiwum Fotogrametrii, Kartografii i Teledetekcji
印刷版ISSN：2083-2214
出版年度：2008
卷号：18
语种：English
出版社：Main Board of Association of Polish Surveyors
摘要：Edge detection is often a fundamental stage of digital image processing in automatic measurement techniques. A number of approaches for edge detection,such as LoG (Laplacian of Gaussian) filtering and Canny-Deriche algorithm,involve using edge-extracting filters. In this paper we present a new edge detection technique. Our approach is based on statistics,specifically on the propagation-separation approach for local likelihood estimation,which was developed in 2006 by J.Pohlzel and V.Spokoiny. This new approach for local estimation involves adaptive determination of a pixel’s neighbourhood,used for estimation of pixel’s intensity. This neighbourhood is determined by a set of weights assigned to pixels in the vicinity of the point of estimation. The value of each weight depends not only on the Euclidean distance between the pixels,but also on a differences in the intensity. The influence of these two factors could be modified by changing the parameters of the procedure. The method,as described briefly here,has been mainly designed for image restoration;however,by using a special set of parameters an effect,similar to segmentation,can be achieved. To obtain the final edge image,it is sufficient to use simply one of the edge extracting filters,for example the Laplacian one. The procedure parameters allow to influence sensitivity of the edge detection. The edge detection results were tested on two fragments of frame images of a city. The edges detected were compared with results of the Canny-Deriche algorithm. The edges obtained were smoother and did not show numerous small breaks. In addition,short,less important edges were less likely to appear. These effects were impossible to avoid using the Canny-Deriche approach.
其他摘要：Detekcja krawĊdzi czĊsto stanowi waĪny etap przetwarzania obrazów cyfrowych w procedurach automatycznego pomiaru. Metody bĊdące obecnie w uĪyciu, w wiĊkszoĞci oparte są na wykorzystaniu filtrów krawĊdziujących. Wymieniü naleĪy tu chociaĪby filtr LoG (Lapl
