基于RTV滤波和改进Otsu优化Canny算子的彩画边缘检测研究

Research on Polychrome Painting Edge Detection Based on RTV Filtering and Improved Otsu-Optimized Canny Operator

摘要: 针对传统算子采用高斯滤波导致的模糊效应、传统非极大值抑制会保留虚假像素点以及人工设定阈值繁琐不稳定等问题，提出基于RTV滤波与可梯度改进Otsu优化的Canny算子改进算法。该算法通过RTV滤波平滑并去除噪声，得到更清晰的图像；并用非线性插值优化非极大值抑制，以梯度幅值加权类间方差最大化原则并采用二次阈值分割策略，消除了虚假边缘像素点，实现高低阈值自适应确定，有效增强了边缘线条的连续性与稳定性。实验结果表明，在BSD500公共数据集上平均F1可达0.836，显著优于其他传统对比方法。

Abstract: To address the blurring effect caused by Gaussian filtering in traditional operators, the retention of false pixels by conventional non-maximum suppression (NMS), and the cumbersome and unstable manual threshold setting, an improved Canny operator algorithm based on RTV filtering and gradient-weighted improved Otsu optimization is proposed. The algorithm uses RTV filtering to smooth and remove noise, yielding clearer images; it further optimizes NMS via nonlinear interpolation, eliminates false edge pixels following the principle of maximizing gradient magnitude-weighted between-class variance, and adopts a two-stage thresholding strategy to adaptively determine high and low thresholds. This effectively enhances the continuity and stability of edge contours. Experimental results demonstrate that the proposed method achieves an average F1-score of 0.836 on the BSD500 public dataset, significantly outperforming other traditional comparative methods.