Abstract: Mean of unsymmetrical trimmed variants is used as a detector for the detection of fixed valued impulse noise is proposed. A fixed 3x3 window is kept constant for the increasing noise densities. The processed pixel is considered as noisy if the absolute difference between processed pixels and mean of unsymmetrical trimmed variants is greater than fixed threshold. Under high noise densities if the processed pixel is noisy and computed median is also noisy then replace the processed pixel with mean of unsymmetrical trimmed variants else if the processed pixel is noisy and computed median is not noisy then the corrupted pixel is replaced by computed median else the pixel is termed uncorrupted and it is left unaltered. If the entire pixels of the current processing window are noisy then global mean of the image is replaced as output. The Proposed Algorithm (PA) is tested on different varying detail images. The proposed algorithm is compared with the standard algorithms and found to give good results both qualitative and quantitatively for increasing noise densities.
K. Vasanth and V. Jawahar Senthilkumar, 2012. A Novel Method for the Removal of High Density Salt and Pepper Noise in Grayscale and Color Images. Asian Journal of Information Technology, 11: 135-143.