N. N. Evtikhiev; S. N. Starikov; M.V. Konnik; R. S. Starikov

Efficient methods of images processing such as edge detection are crucial for correlation objects recognition. This paper deals with images contouring algorithms for edge detection on images that were taken in difficult weather conditions. It is shown that proper application of specialised contouring methods can improve robustness and accuracy of objects recognition dramatically. Obtaining contoured images that are noise-free and tolerant to weather conditions is the main problem in such tasks. Such algorithms must be noise-tolerant and provide contoured images that are independent from conditions in which pictures are taken. Furthermore, such algorithms must be fast enough to be implemented as a programs for airborne systems. Such circumstances demand application of sophisticated algorithms of edge detection. A comparison of edge detection algorithms for contouring of aero photography images taken in difficult weather conditions is presented in this work. There were compared algorithms of first-derivative (Sobel and Canny), second-derivative (Laplassian of Gaussian), and morphological contouring algorithm that have been developed. Results of edge detection on test images by all observed algorithms are provided. Estimations of performance as well as accuracy of such contouring algorithms are performed. It is shown that the morphological edge detection algorithm exhibits greater robustness of objects recognition on contoured images. The accuracy of the morphological edge detection algorithm is better on 8\% than Sobel, on 16\% than Canny, and on 6\% than LoG in average by correlation criterion. The performance of the morphological algorithm is provided as well. The morphological algorithm is 2 times faster that Canny on contouring images of 400х400 pixels, and 1.8 times faster on contouring images of 800х800 pixels. On the other hand, the morphological algorithm is 3.6 slower than Sobel and LoG on contouring images of 400х400 pixels, and 2 times slower on contouring images of 800х800 pixels. Nonetheless, the morphological algorithm demonstrates better contouring accuracy that makes the morphological algorithm an optimal solution for such kind of tasks. Obtained results allow us to say that the morphological algorithm, which is presented in this paper, is an optimal trade-off between speed and accuracy on contouring images taken in difficult weather conditions

1. Sobel I., Feldman G. A 3x3 isotropic gradient operator for image processing. Never published but presented at a talk at the Stanford Artificial Project. 1986.
2. Canny J. A computational approach to edge detection // IEEE Transactions on Pattern Analysis and Machine Intelligence. 1986. No 6. PAMI-8. Р. 679-698.
3. Verri A., Trucco E. Introductory technics for 3-d computer vision. Prentice Hall. 1998.
4. Jean Serra. Image Analysis and Mathematical Morphology. London: Academic Press. 1982.
5. Maragos P., Schafer R.W. Morphological filters. Part i: their set theoretic analysis and relations to linear shift invariant filters // IEEE Trans. Acoust., Speech, Signal Process. 1987. V. ASSP-35. N. 8. Р.1153-1169.
6. Peters R.A. A new algorithm for image noise reduction using mathematical morphology // IEEE Transactions on Image Processing. 1995. V. 4. N. 3. Р. 554 - 568.
7. Вудс Р., Гонсалес Р. Цифровая обработка изображений. М.: Техносфера. 2006. 1072 с.
8. Евтихиев Н.Н., Конник М.В., Стариков Р.С. Разработка алгоритмов выделения контуров для оптико-электронного корреляционного различения изображений // Сборник трудов Научной сессии МИФИ-2007. 2007. Т.15. С.137-138.
9. Otsu N. A threshold selection method from gray-level histograms // IEEE Trans. Systems Man Cybernet. 1979. V. 9(1). Р.62 - 66.