Abstract: Improving correction detection,meanwhile decreasing miss detection and false alarm is a challenge in random-valued impulse noise detection.A two stages noise detection algorithm was proposed,at first stage,noisy pixels were detected when their pixel-value density was less than certain threshold,this stage included five iterations,noise pixels were removed through median filtering in every iteration,and the filtered images were the input images for next iterations.At second stage,noisy pixels were detected through four directions,and adaptive threshold was based on MAD.The test images were 512×512 Lena and Boat added by random-valued impulse noise from 10% to 50%.Simulations indicate that with the increase of noise density,false alarm remains low and miss detection remains theoretical low level.

Key words: Image denoising,Random-valued impulse noise,Noise detection,Pixel-value density