Research On Rank-Ordered Statistic Based Random-Valued Impulse Noise Detection

Random-valued impulse noise is a kind of impulse noise,which is hard to remove by ordinary methods for its characteristics of distribution and amplitude,especially when the noise level is high.For now,detecting impulse noise before filtering is an effective way.The impulse detector is used to identify noise and noise-free pixels in the image.By using noise detector,we can mark noise pixels efficiently,and only process these pixels instead of all pixels.This detect-and-remove procedure can remove most noise in the images,as well as preserve most details of the image.In this paper,a new idea of noise detection is proposed based on rank-ordered statistics.The idea is to transform the distance between current pixel and reference pixels by a piece-wise power function,and then the transformed distance is ranked in increasing order.After that,the small half of these distances are added as the proposed statistics.The statistics represent the possibility of being noisy.We can set an adaptive threshold according to the ratio of noise,and mark all pixels with higher values than the threshold as noise.The piece-wise power function is the main innovative idea in the paper.The distance between statistic values of noise and noise-free are enlarged by increase the absolute differences in different degrees,in order to identify impulse noise effectively and precisely.In the stage of denoising,the proposed noise detector are combined with a impulse noise removal algorithm as a complete detect and remove procedure.The procedure is performed in an iterative way,and the initial threshold of impulse detector is obtained by an adaptive function.During the iterations,the parameters of distance transform function are changed to obtain better detection results.A lot of experimental results indicate that the proposed impulse detection method has better detection results than most present detection methods.Further more,the denoising algorithm which is based on the proposed detection method receives good performances both in subjective and objective ways.