The circle detection problem focuses on finding all circle shapes within a given image. In fact, circle detection has several applications in real-life problems arising in agriculture, ophthalmology, and oceanography, among others. Despite many approaches having been proposed to deal with this problem, our work is motivated by two main issues: (1) the limitation of a recently proposed evolutionary algorithm and (2) the lack of benchmark images to fairly compare current approaches. To address the first issue, we introduce an effective evolutionary algorithm with a pre-processing noise reduction step. The proposed evolutionary algorithm’s goal is to match several randomly generated circles with a point cloud extracted from an edge map of the original image. These circles are individuals in the population where the fittest one in the last generation is a detected circle. Henceforth, by removing the points corresponding to such circle and repeating the process, all circles within the image can be detected. We propose and make publicly available a set of synthetic, hand-drawn, and real images with different features to address the second issue. To assess our approach’s performance, we apply it to the set of proposed images that include challenging features. Experimental results show that our method is competitive compared with the well-known Circle Hough Transform and as well as with EDCircles.

圆形检测问题侧重于在给定图像中查找所有圆形。事实上，圆检测在农业、眼科和海洋学等领域出现的现实问题中有多种应用。尽管已经提出了许多方法来解决这个问题，但我们的工作受到两个主要问题的推动：（1）最近提出的进化算法的局限性和（2）缺乏基准图像来公平比较当前的方法。为了解决第一个问题，我们引入了一种具有预处理降噪步骤的有效进化算法。提出的进化算法的目标是将几个随机生成的圆与从原始图像的边缘图中提取的点云进行匹配。这些圆圈是群体中最后一代中最适合的个体是检测到的圆圈。以后，通过去除与该圆对应的点并重复该过程，可以检测图像内的所有圆。我们提出并公开一组具有不同特征的合成、手绘和真实图像，以解决第二个问题。为了评估我们的方法的性能，我们将其应用于包含具有挑战性特征的建议图像集。实验结果表明，与著名的圆形霍夫变换以及 EDCircles 相比，我们的方法具有竞争力。我们提出并公开一组具有不同特征的合成、手绘和真实图像，以解决第二个问题。为了评估我们的方法的性能，我们将其应用于包含具有挑战性特征的建议图像集。实验结果表明，与著名的圆形霍夫变换以及 EDCircles 相比，我们的方法具有竞争力。我们提出并公开一组具有不同特征的合成、手绘和真实图像，以解决第二个问题。为了评估我们的方法的性能，我们将其应用于包含具有挑战性特征的建议图像集。实验结果表明，与著名的圆形霍夫变换以及 EDCircles 相比，我们的方法具有竞争力。