Research On The Algorithm To Remove Impulse Noise In Digital Image

Images are usually contaminated by all kinds of noise when it is transmitted, collected, and received due to poor quality of sensor components or environment factors and so on in real life. Impulse noise is one of the noise which will degrade the quality of images, so the impulse noise pixels must be removed by filtering before image enhancement, image segmentation, feature extraction. Some new Corresponding algorithms to remove impulse noises are then presented based on the gray values and the distribution modes of the noise pixels after analyzing the five common impulse noise models, and the main research of this thesis can be summarized as follows:(1) One algorithm by replacing the gray value of the noise pixel in filtering window with the median of signal pixels in edge of filtering window, the other algorithm realize the removal of salt and pepper noise pixels with two steps of mean value calculation for signal pixels, Simulation can demonstrate these two algorithms not only perform well in removing impulse noise in images, but also achieve better filtering result in a shorter time than other excellent algorithms in nowadays.(2) The algorithm aimed for random values impulse noise composed by the same range length of low noise and high noise is firstly utilized to judge whether the pixel is suspected noise pixel with the signal pixels in current filtering window, then a set composed of pixels from pozidriv section from current filtering window is to determine whether the suspected noise pixel is really a noise pixel. The results of experiment on this algorithm show that the referred algorithm have a good performance in denoising the impulse noise, especially for the original image with flat homogeneous region.(3) As for the most sophisticated impulse noise model, the new algorithm in this thesis can seize the two boundaries b1and b2to determine whether the current detecting pixel is noise pixel by creating a histogram difference matrix and calculating the average number of gray scale in histogram at first, then the noise pixel which owns more signal pixels around itself in filtering window have a priority to be filtered and its gray value is replaced by the median of signal pixels referred above, besides, this noise pixel should also be regarded as signal pixel after filtering.