No abstract

Different digital applications often rely on methods which allow the definition of the edge of the objects in an image. The edge is expressed as a discontinuity of grayscale in the image and can have significant information about the objects in the image. Thus, edge detection can be useful for a wide variety of purposes such as object square calculation and object shape recognition in Earth-remote sensing (ERS), separating an object from the background in computer vision applications, military applications including target recognition and traffic analysis, and security applications including data encryption and watermarks. All of this requires significant edge detection accuracy, which is correlated with visual edge appearance. That was the reason to perform the comparative analysis for the different edge detection techniques for ERS images in particular. This study consists of two approaches for edge detection, specifically, the first and second derivative technique. Assessment of the edge detection methods is performed in terms of PSNR and SSIM metrics, which definitely are the widespread and reliable metrics for quality assessments purposes, in comparison with visually ideal appeared edge in each of the satellite image. Selected ERS images have different degrees of detailing, resolution and object shape. As a result, of the current study we can state that binary thresholds for edge detection should be chosen in accordance with a compromise between false detection and miss detections for every method. In particular, the second derivative method gives fewer missing edges, but for the same Roberts threshold it gives false edges. The first derivative method in turn is a more rough algorithm, which misses edges, especially with images of high detailing.

No abstract