Switching vector median filters based on non-causal linear prediction for detection of impulse noise

A new group of switching vector filters based on the non-causal linear prediction for the detection of impulse noise from colour images is presented. The proposed filters utilise the non-causal linear prediction coefficients obtained from the block-by-block autocorrelation function to find prediction vector pixel value at the centre of the filter window. Thirteen prediction coefficients are selected from the autocorrelation matrix obtained from a block of an image, and these coefficients are used to predict all pixels in that block. The difference between the predicted pixel and the original decides whether the pixel is corrupted with impulse noise. Extensive simulation experiments indicate that the new vector filters outperform the other vector filters currently used to eliminate impulse noise from colour images.