Selective vector median filtering of colour images

The vector median filter (VMF) has proved to be a highly effective tool for removing impulse noise from colour images. However, when images are highly corrupted, or when high fidelity is required, the application of a vector median filter may lead to unacceptable levels of colour image distortion. We present a development of the VMF which divides the filtering process into two distinct stages: noise detection; and noise removal. This allows selective application of the VMF to only those neighbourhoods which contain "noisy pixels" and subsequently allows these pixels to be excluded from the calculation of the vector median. It is therefore possible to substantially improve the performance of the VMF when operating on highly corrupted images or when an application requires a high level of fidelity.