Random-valued impulse noise removal based on component-wise noise detector with auto-tuning function and vector median interpolation

We present a novel switching median filter for the removal of random-valued impulse noise on a color image. The present filter consists of a new noise detector with auto-tuning function, and a vector type noise remover avoiding the occurrence of the color artifacts after the noise removal. The detector has a parameter, which is automatically tuned only by using the distribution information of the noise signal without any training signals, and detects the pixel corrupted by the random-valued impulse noise with the parameter. And the detector operates in each channel independently. As a noise remover, the ordinary vector median filter is employed to interpolate only the pixel which is detected as noisy one by the detector. Through the experiments using some digital color images, the validity and the effectiveness of the present method are verified.