由于多种原因 ,部分SeaWiFS卫星图像数据中存在比较严重的椒盐噪声。该文在分析SeaWiFS椒盐噪声特征的基础上提出一种基于窗口内均值与均方差比值序列差值的统计比值差值排序滤波器 (StatisticalRatioRankOrderedDifferencesFilter,SRROD) ,并讨论如何使用该滤波器技术有效地对椒盐噪声进行白点噪声、黑点噪声检测和消除。与常用的中值滤波和其他滤波器比较 ,该方法能在有效消除椒盐噪声的同时 ,保持图像数据中其他位置的点不受影响。通过灵活地调整不同的阈值可以获得不同的滤波效果。最后 ,讨论了如何从有效峰值信噪比 (EffectivePeakSignalNoiseRatio,EPSNR)分布图上提取最优阈值对的方法。

Due to some uncertain reasons, many seaWiFS satellite images are corrupted by impulse noise. In this paper, we firstly analyzed the characteristics of impulse noise and proposed a new rank ordered filter based on the difference of sequence of mean and standard deviation ratio, which is named as Statistical Ratio Rank Ordered Differences Filter (SRROD filter). Second, We described the impulse noise detection and removal algorithm in detail. Similar to traditional median filter, the processing of SRROD filter is implemented by a moving window concerning to different size of neighborhood. Compared with median filter and other existing filters, our filter could effectively remove impulse noises while preserving other valid pixels without or only with little modification, with the cost of about 10 times extra computing time than median filter. To better assess the noise removal quality, we have derived a more reasonable variable to estimate the image quality. That was the Effective Peak Signal to Noise Ratio( EPSNR ), instead of the traditional Peak Signal to Noise Ratio (PSNR) . The estimation of EPSNR also showed that much better improvement has been achieved with our algorithm than median filter. In our algorithm, through controlling the value of lower and upper threshold, different filter effect could be achieved. One of the key to successfully remove impulse noise is the way to choose an optimal threshold pair. Thus we also made fully discussion of finding an optimal threshold pair. Based on the estimation and assessment for the distribution map of the EPSNR according to different lower and upper threshold pairs, a nearly optimal threshold could be found. The Laplacian transformation was found very useful in finding this optimal threshold pair. The estimated optimal threshold pair was applyied to a full scene SeaWiFS image(Channel 2)and obtained a fairly good result, in which the result was also shown in our paper. Finally, some concluding remarks and limitations of our algorithm as well as the suggestions are given. The further work to be conducted also presented in the conclusion section.