An Adaptive Random-valued Impulse Noise Removal Algorithm Using A Two-phase Detector

During the acquisition and transmission processes,images can be corrupted by impulse noise due to malfunctioning pixel elements in the camera sensors and bit errors in analog-to-digital conversions.Image denoising is a crucial step because contaminated images severely hinder many operations,including edge detection,image segmentation and object recognition.This paper proposes a new algorithm for the detection of random-valued impulse noise(RVIN).First,a two-phase noise detector is developed to identify noise candidates.In the first phase,we calculate the gray-value distances separately between the current pixel and its neighbors aligned with four main directions.A piecewise power function is applied to the distances in order to amplify the differences between corrupted pixels and uncorrupted pixels.In the second phase,we calculate the number of similar pixels in the local window and use this information to further refine the detection performance.Using a piecewise power function to transform the directional absolute differences and performing two-phase noise detection is one of our innovations.Different from previous studies that using a fixed detection threshold,we use the statistical characteristics within the local window,a variation of the median of the absolute differences from the median,to adaptively determine the detection threshold.Compared with other noise detectors,the proposed two-phase detector can effectively reduce the rate of false detection.By combining the two-phase noise detector with an improved edge-preserving regularization filter,an iterative denoising procedure is proposed for removing RVIN.An effective adaptive stop criterion is also developed based on the number of filtered pixels for the detection based iterative denoising procedure.Extensive experiments show that the developed denoising algorithm achieves good results in terms of quantitative evaluation and visual quality.