鉴于开关中值滤波在椒盐噪声检测和去除方面的应用合理性, 本文分别设计实现了基于信号局部差异性和基于信号方向差异性的椒盐噪声检测算法。这两种算法均属于二级噪声检测方法, 且第一级检测手段都是基于灰度范围准则。两种算法的不同点主要体现在第二级检测算法上, 前者基于局部差别准则, 后者基于方向差别准则。在方法评价部分, 首先通过分析和实验确定两种算法的最优参数设置; 然后通过对不同噪声密度的测试图像去噪来评价两种算法的去噪效果。结果表明: 基于方向差异性的算法比基于局部差异性的算法具有更好的性能, 且两种算法的去噪效果都与噪声密度成反比。需要注意的是, 这两种算法都容易将图像中的细微边缘或细节像素误判为噪声点, 即在噪声的检测过程中, 只能避免对图像中主要边缘和轮廓像素的误判, 还无法对图像中的细微边缘和细节进行精确判定, 这也是开关二级噪声滤波算法今后的主要改进方向。另外, 算法效率测试结果表明两种算法具有相似的计算时间, 从而验证了两者之间的算法结构相似性。

Switch median filters are very suitable for detection and removal of the salt and pepper noise. An algorithm based on signal local difference and an algorithm based on signal directional difference for detecting salt and pepper noise were designed in this article. Both of them belong to two-stage noise detection algorithms, and the first-stage detection for both is based on gray-scale range criterion. The difference between the two algorithms embodies in the second-stage detection. The former is based on local difference criterion, and the later is based on directional difference criterion. In the section of algorithm evaluation, first, the optimal parameter settings for the two algorithms were determined through analysis and experiments. Then, the denoising effects of the two algorithms were evaluated by denoising the test images with different noise density. The evaluation results show that the algorithm based on directional difference has better performance than the algorithm based on local difference. For both of the two algorithms, the denoising effects are inversely proportional to noise density. But more importantly, the two algorithms may be prone to misjudge the image-edge pixels or detail pixels as noise points. This will be the main improvement direction of the switch median filtering algorithm in the future. Finally, the two algorithms have the similar computational efficiency. It verifies the structural similarity between them.