线性滤波算法的机器视觉适用性研究

Research on the Machine Vision Applicability of Linear Filtering Algorithm

摘要: 在机器视觉系统实际工作过程中，由于状态噪声的多样性，很难实时选择出最佳滤波算法。论文针对线性滤波算法：时域递归滤波、高斯滤波、高斯–拉普拉斯滤波、二项式系数滤波的算法模型、计算速度、输出结果进行了深入研究，通过迭加随机噪声组合测试了上述算法的性能，分析了其在工程实际中的适用性。实验结果表明，在迭加白噪声与高斯噪声时，高斯滤波实现了时间与精度的最佳折中；在迭加椒盐噪声时，高斯–拉普拉斯滤波明显优于其它三种算法，为机器视觉系统优化图像处理效果提供了理论和技术支持。

Abstract: In the actual working process of machine vision system, due to the diversity of state noise, it is difficult to choose the best filtering algorithm in real time. In this paper, the linear filtering algorithm: Instant Domain Recursive filtering (IDR), Gaussian filtering, Laplace of Gaussian filtering (LOG), Binomial Coefficient filtering (BC) algorithm model, calculation speed and output results are studied in depth. The performance of the above algorithm is tested by superposition random noise combination, and its applicability in engineering practice is analyzed. The experimental results show that Gaussian filtering achieves the best compromise between time and accuracy when white noise and Gaussian noise are superimposed; when salt and pepper noise are superimposed, Gaussian Laplace filtering is obviously superior to the other three algorithms, providing theoretical and technical support for the optimization of image processing effect of machine vision system.