A Study Of The Algorithm For Median Filter Based On The Analysis Of Grey Absolute Relation

With the continuous development of multimedia and communication technologies，themeasurement information of image signals are more widely used in various fields. However,the image of the visual effects in the generation, transmission and recording process are oftenaffected by various noise. And it also brought a lot of inconvenience for edge detection,image segmentation, feature extraction, pattern recognition and subsequentprocessing.Therefore, image denoising algorithm before these processes is a very importantprocessing step to reduce noise. A variety of image noise can be obtained in real life, amongwhich impulse noise is one of the most common forms.According to the characteristics of the actual image and the noise distributioncharacteristics, various denoising methods are proposed. These methods can be divided intotwo categories, namely spatial filtering and transform domain filtering method. With thedevelopment of image processing technology, a variety of filtering algorithm are put forwardto filter image noise on the basis of keeping image detail. These methods have a differentrange of applications and different effect for the filter effect.So it is very difficult to decidewhich one is the best. In this paper, a lot of denoising method in the previous imageprocessing were studied. Aimed at the characteristics of the gray image, the absolute grayrelational analysis method of gray system theory was applied for image analysis and noisedetection, which not only improved the rapid development of denoising and provided a newmeans of disposing for denoising technique,but also expanded the application fields of graysystem theory in image denoising area.This paper first described the research background, the purpose of topics and researchstatus at home and abroad.The research contents through understand denoising in imageprocessing was expanded correspondingly. Then, the types of noise and evaluation criteria ofimage quality were described, the research methods of the traditional denosing and a range ofmethods for improving image quality were discussed. Correspondingly, the gray systemtheory was introduced. The basic theory of Deng’s gray relational analysis and model features was described and the related idea of gray correlation analysis was introduced. On the base oftraditional filter advantages and characteristics of impulse noise, a new filtering algorithmwas improved by gray correlation analysis method. This method is better for filtering noise,which can keep the image detail information.This research work is mainly concentrated in the following respects:(1) Described the noise model and the evaluation criteria of image quality and introducedseveral classic spatial domain and transform domain denoising method. Median filteringalgorithm based on gray correlation analysis was improved for the detail protection andclarity of image denoising.(2) Introduced the concept of gray system，described the significance of the grayrelational analysis, judged Deng’s gray relational grade of the main ideas and calculations，andpointed out the inadequacies in the application. On this basis, the gray correlation analysistheory was introduced， and the method of calculation was described. Using the gray featuresof images and image noise， the gray absolute correlation analysis method was adopted toincrease noise image. The result shows this method could better distinguish the point of thesignal and noise pixels in the image， and improve the quality of image denoising.(3) Used gray correlation analysis method for the detection of image noise， andcombined with the advantages of the median filter for image filtering. First, an n×n template(n is an odd number and greater than3or equal) was constructed. Then, the pixels waiting ofimage noise was treated as the position center and the pixels in the template were divided intotwo groups of sequence. Finally, the similarity of these two groups of sequences werecalculated through the gray correlation analysis. According to the threshold, the current pixelwas judged whether it’s the noise. After detecting noise points， pixels in an image weredivided into two types, namely noise pixels and non-noise pixels. For non-noise pixels, theoriginal value is reserved.As for the noise point pixel, the median values of neighborhoodpixel were selected to discriminate whether the currently selected value points was the noise.If the median value was noise pixels, it would be released,search the non-noise pixels fromneighborhood and put the median value of these non-noise pixels as the final pixel to beoutput。 In order to verify the validity of the algorithm, the images with low-density noise andimages with high-density noise were filtered under the programming environment ofVC++6.0. Filtering results were compared between the standard median filter and extremevalue median filter from two aspects of filtered visual effects and objective performanceevaluation criteria. Simulation experiment results show the algorithm can effectively removethe noise and keep image detail to some extent. Especially in the case of larger noise density,the algorithm has better advantages than any other algorithm.