Second order differential hyperbolic tangent based gradient estimation to improve Canny edge detection in digital images

Canny algorithm is an optimal algorithm to detect edges in images. Canny edge detector (CED) commonly uses the differential of Gaussian (DoG) filters to estimate the edge information. Although, DoG provides better tradeoff between smoothing level a and edge detection, its smoothness affect the performances of edge detection (ED). In particular, when the smoothing parameter is greater than one, the localization, and detection performances are degraded. This proposed approach makes use of a second order differential hyperbolic tangent filter to determine the gradients, and provides well-localized edge features than DoG filter. Mean squared error (MSE) and Figure of Merit 'F' parameters are used to measure the detection and localization performances. This algorithm is tested on 44 different natural images, and improves the 'F' to 0.9787 and reduces the average MSE to 0.2752 while the conventional CED has 'F' as 0.9422 and MSE as 0.2905. © 2012 ACECR.