The artificial neural network for image contour extraction

The solutions of some modern applied image processing problems require the use intellectual image analysis tools. We deal with one of such problems-processing of cosmic and aerial images. The possibilities of automatic interpretation of the aerial and cosmic images of the Earth surface are restricted by natural and diverse factors: seasonal, daily and other similar to image variations and geometric transformations. In automatic processing of images with complicated distortions, the traditional correlation-based image processing methods fail when dealing with the mentioned image distortions. It is shown that the analysis of curve forms which correspond to object contours on the image, is the most suitable tool for solving this problem. The neural algorithm presented was applied to contour extraction on real LANDSAT images.