去除图像中的椒盐噪声可转化为二维曲面的重建问题。选用Multi-Quadric函数对图像中损失的信息构造插值格式,自动选择待插值点和插值参考点并求解插值方程组得到处理后的图像。实验证明,本方法可以在很少破坏图像细节的情况下去除大部分甚至全部噪声,并且在噪声密度非常大的情况下仍然可以还原相当多的图像信息。对噪声密度为50%和90%的单色Lena图像进行处理,该方法得到的信噪比比自适应中值滤波高6dB以上。

Removing salt and pepper noises in digital images can be converted to reconstruction of 2-dimensional surface. This method selects multi-quadric function to build interpolation of missing information in images, automatically selects interpolation points and referenced points, and solves the interpolation equations to get the processed images. The experiments shows that this method is able to remove almost, even all noises with little damage to details of images. Also, it is able to restore more information when the noise rate is very high. While processing the Lena images with 50% and 90% salt and pepper noises, the result of this method is 6 dB higher than which of adaptive median filter in SNR.