Edge detection for Very High Resolution satellite imagery based on Cellular Neural Network

In the context of Very High Resolution (VHR) satellite imagery, edge detection is one of the most important and difficult steps in image processing and pattern recognition. This paper presents the use of Cellular Neural Network (CNN) in edge detection and shows its capacity to locate and identify discontinuities in the gray levels of the objects that appear in VHR image. The results show that the characteristics of the CNN, in terms of its local connectivity, are those that allow greater extraction of continuous edges. The multi-layer processing structure design of the CNN allows to identify a definitive edge in urban environment. To evaluate the results, a metric of peak signal to noise ratio (PSNR) has been introduced as a manner to rank the accuracy of the resultant edge determined by the assessed methods. The extracted VHR features with the CNN edge detector include accuracy of edge location and better linking of edge segments.