Abstract
The existing edge detection methods can often encounter several common shortcomings: discontinuous edges, weak edges, easy to be affected by noise, and difficulty in setting gradient thresholds. To address these shortcomings, this paper proposes a neural-like computing model, called dynamic threshold neural P systems with orientation, termed as ODTNP systems. In addition to the spiking and dynamic threshold mechanisms, ODTNP systems also integrate the gradient magnitude and gradient direction information. The combination of these mechanisms is beneficial to ODTNP systems to achieve edge detection. Based on ODTNP systems, a novel edge detector is developed. The proposed ODTNP-based edge detector is evaluated on the benchmark Berkeley segmentation data set (BSDS500) and is compared with five baseline edge detection methods and four state-of-the-art deep-learning-based edge detection methods. Experimental results demonstrate the availability and effectiveness of the proposed edge detector.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have