This study aims to present an effective method to detect ellipses on images with low signal-to-noise ratios (SNR). Firstly, we analyze four major interferences caused by low SNR images. A nonlinear circumference-based ellipse detection method is proposed, which uses a nonlinear circumference to vote on the parameter space constructed from input images and a spatial hierarchical search strategy to find the optimum ellipse parameters. The nonlinear circumference is designed to describe the possibility that selected edge points will generate a given ellipse. Experiments on six low SNR datasets show that our method outperforms five existing methods in terms of recall, precision and F-measure. Furthermore, a definition based on the nonlinear circumference is proposed to quantify SNR of synthetic images with ellipses. Experimental results demonstrate that our method can detect ellipses on images close to 0.2 dB while state-of-the-art methods can almost only reach 1.2 dB.

中文翻译：

---

低信噪比图像中基于非线性圆周的鲁棒椭圆检测

本研究旨在提出一种有效的方法来检测低信噪比（SNR）图像上的椭圆。首先，我们分析低信噪比图像造成的四种主要干扰。提出了一种基于非线性圆周的椭圆检测方法，该方法使用非线性圆周对由输入图像构造的参数空间进行投票，并使用空间分层搜索策略来寻找最佳椭圆参数。非线性圆周旨在描述所选边缘点生成给定椭圆的可能性。对六个低 SNR 数据集的实验表明，我们的方法在召回率、精度和 F 测量方面优于现有的五种方法。此外，提出了基于非线性周长的定义来量化椭圆合成图像的信噪比。实验结果表明，我们的方法可以检测图像上接近 0.2 dB 的椭圆，而最先进的方法几乎只能达到 1.2 dB。