복합잡음 제거를 위한 디지털 합성 필터에 관한 연구

Abstract

Most of the image processing and computer vision systems extract useful information from image images. However, actual image images tend to cause noise due to various reasons in the system in a process of acquisition, transmission, and saving. Noise removing is of an inevitable pre-conditioning procedure in all image processing areas. Therefore, studies have been actively conducted to effectively remove noise added in the image. There are many types of noises added in the image depending on the causes and forms. As for representative examples, there are salt & pepper noise, AWGN(adaptive white Gaussian noise), and mixed noise of them. As for the representative filter for removing salt and pepper noise, there is a median filter. As a representative non-linear filter, median filter arranges pixels in the local mask in an ascending order and selects the median value. This filter features with an outstanding function of removing noise in the low density of salt and pepper noise. However, it is insufficient in removing noise and preserving edges in the high noise density area. Average filer is to remove AWGN. Average filter is a representative linear filter calculating the average value of pixels in the local mask and smoothing image images for the process. This filter represents an outstanding function of removing noise in the low frequency with fewer changes in brightness of image images. However, it is unable to consider detailed information such as edges in the high frequency with a wide range of changes in brightness. However, existing filters tend to have a deteriorated function of removing noises depending on the types of noise and entered image images due to limitations of filter. In this study, noises were processed by dividing into salt and pepper noise and AWGN in order to effectively remove mixed noise added in the image image. If center pixel in the local mask was damaged by salt and pepper noise, final outcome of edge preserving filter, local histogram weight filter, and median filter was synthesized for the process. If it was damaged by AWGN, algorithm has been suggested that final outcome of weight filter in the use of standard deviation in distance changes, pixel changes, and local mask was synthesized and processed. In addition, PSNR was used to prove the outstanding feature of suggested algorithm comparing the functions of it with the ones of other existing methods.