针对椒盐噪声的特点, 提出了一种以回归型最小二乘支持向量机(Least square support vector regression , LS-SVR)为数据恢复算法的开关型滤波器。首先利用max-min算子对滤波窗口中心点进行噪声判别, 若中心点不是窗口极值, 则将其正常输出, 若为极值, 则将其判定为噪声, 并进一步将窗口分为只有中心点被污染和多点被沾染二类, 利用LS-SVR良好的数据逼近能力, 对窗口进行曲面拟合, 实现了被污染的数据点的有效恢复, 减小了被误判为噪声的数据点的损害。为提高算法的运算速度, 根据滤波策略和LS-SVR的特点, 先期构造了二种LS-SVR卷积模板, 将LS-SVR的训练过程转化为了简单的加权求和运算, 增加了算法的实用性。实验表明, 这种方法具有较好的细节保护能力和较强的噪声去除能力。

Aiming at the characteristic of salt & pepper noise, the switch salt & pepper noise suppression algorithm for noise points data restoration based on LS-SVR (least square support vector machines regression) is presented. First, the max-min operator is used to differentiate the noise at the filter windows center, if the center is not the windows extremum, they will output naturally, otherwise they will be judged as noise and divide windows into center corrupted points and multi-corrupted points. The well data approach ability is utilized to fit the curved surface for windows. It realized efficient restoration of data corrupted and reduced the data points damage for miscarriage of justice as noise. To improve the arithmetic operation speed, it constructed the two type LS-SVR convolution mask based on the filtering strategy and LS-SVR characteristic, training process of LS-SVR is turned into simple weighted summation operation which increased the algorithmic practicability. Experiments show that the proposed algorithm has better detailed protecting ability and better removing noise ability.