Improved active contour model for satellite images

Abstract

Snakes or active contours are used extensively in computer vision and image processing applications, particularly to locate object boundaries. The traditional snake was sensitive to initialization of contour; also one snake was able to detect one object only. In presented research work we have developed improved active contour model for detection of multiple objects well as edge detection in satellite images. We have proposed modified gradient vector flow as external force to make edge detection insensitive to initialization and to exact contour on edges. Method of capturing of relevant control points, so as to neglecting extra control points is introduced to get proper edges. For satellite images new techniques as pre processing to enhance edges by removing noise, double thresholding based method for deleting of excess control points, and average based thresholding for obtaining continuous edges by eliminating need of complex interpolation is developed. The algorithm is tested on variety of images and cases. Both internal and external initialization of contours gives satisfactory edges. Snake work efficiently for both noise free as well as noisy images. Algorithm also outperforms in terms of time complexity as compare to other edge detection algorithm such as canny edge detector.