Edge detection robust to intensity inhomogeneity: A 7T MRI case study

Abstract

Edge detection is a fundamental operation for computer vision and image processing applications. As of 1986, John Canny proposed a methodology that became known due to its simplicity, small number of parameters, and high accuracy. The method was designed to optimally detect, locate, and trace single edges over each local gradient maximum. Since then, a number of works were proposed but none of these improvements were capable of dealing with non-uniform intensity, which are notably present in ultra high field magnetic resonance imaging (MRI). In this paper, we evaluate the effects of inhomogeneity correction over automatic edge detection methods over 7T MRI. Importantly, we propose a non-supervised edge detection method which improves the accuracy of state of the art in 28.0% as detecting head and brain edges.