The Canny algorithm is one of the most commonly used edge detectors due to its superior performance, especially in noisy environments. Its main limitation is that it is time consuming due to its multistage nature and the use of complex computational operations, primarily hysteresis thresholding. For this reason, many efficient implementations of the Canny edge detector have been developed on different accelerating platforms, such as ASICs, FPGAs and GPUs. The two main limitations of the GPU implementations developed to date are the bottleneck caused by the hysteresis process, and the use of fixed hysteresis thresholds. To overcome these issues, a novel GPU-based unsupervised and distributed Canny edge detector is proposed in this paper. Experimental evaluation showed that our Canny edge detector fully satisfies real time requirements, as it only requires 0.35 ms on average to detect edges on 512\(\times\)512 images, and that it is faster than existing GPU and FPGA implementations.

Canny 算法因其卓越的性能而成为最常用的边缘检测器之一，尤其是在嘈杂的环境中。它的主要限制是由于其多级性质和使用复杂的计算操作（主要是滞后阈值），它非常耗时。出于这个原因，Canny 边缘检测器的许多有效实现已经在不同的加速平台上开发，例如 ASIC、FPGA 和 GPU。迄今为止开发的 GPU 实现的两个主要限制是滞后过程引起的瓶颈，以及使用固定滞后阈值。为了克服这些问题，本文提出了一种新型的基于 GPU 的无监督分布式 Canny 边缘检测器。实验评估表明，我们的 Canny 边缘检测器完全满足实时要求，\(\times\) 512 张图像，并且它比现有的 GPU 和 FPGA 实现更快。