Computational methods in stochastic modeling for data mining

Abstract

In image processing, various linear and non linear filtering newlinemethods have been proposed for the removal of impulse noise. Linear newlinefiltering techniques used for noise reduction in images are simply given newlineby the average of the pixels contained in the neighborhood of the filter newlinemask. However, linear filters cannot effectively reduce impulse noise and newlinehave a tendency to blur the edges of an image. In such situations, newlinemedian filters, which are non linear filters, provide an effective solution. newlineCompared with convolution filters, the median filter is more robust in newlinethat a single very unrepresentative pixel in the filter window will not newlineaffect the median value significantly. Also, since the median must newlineactually be one of the pixels in the filter window, the median filter does newlinenot create new pixel values when the filter crosses an edge. For this newlinereason, the median filter is better in preserving sharp discontinuities newlinethan linear filters. Unfortunately, the median filter is prone to alter pixels newlineundisturbed by noise, thereby causing a number of artifacts including newlineedge jitter and streaking. Modified forms of the median filter which still newlineretain the rank order structure have been proposed to overcome these newlineshortcomings. Basically, the task is to decide when to apply the median newlinefilter and when to keep the pixels unchanged. Among those are the newline27 newlineCenter-Weighted Median filters, which give current pixel a large weight newlineand the final output is chosen between the median and the current pixel newlinevalue, detail-preserving median filters and rank ordered mean filter newlineexcludes the current pixel itself from the median filter, progressive newlineswitching median filter, soft-decision-based filter and prediction-basedfilter. newline