Edge detection in non-linear scalable space

Abstract

Automatic image processing and computer vision applications often include object extraction and recognition. To extract object boundaries better, high-quality methods of edge detection are required. Objects in real-world photos usually have complex structure, and discovering their edges with higher precision is a very important and tough task. Better edge detection drastically increases the quality of the entire process. Unfortunately, edge detection problem still has no satisfactory solution in the case of real-world color images, because efficient approaches exist for grayscale images only. Known edge detection techniques usually either consist of special filters applied to a grayscale image, or implement rather unpredictably behaving artificial intelligence solutions. We propose a new powerful mathematical tool to extract edge information not only from intensity but from color distribution as well. Correctness of edge detection gets better by means of an adaptive mathematical model of non-linearly scalable color space. In our research, we consider flexible non-linear transformation of the color space with a few parameters. It helps reliably extracting otherwise faded edges delimiting the objects. We compare the results of edge detection process based on this new technique to well-known base-line approaches of Canny, Prewitt, Sobel and Wallace filters.