Study Of Blurred Measurement For Image Based On Multi-scale Analysis

Images have been widely used in many industries, such as security, aerospace, machinery,medicine and so on, so it plays an increasingly important role. However, images will losesome information during the process, which results the loss of image quality. It is moreimportant to recognize the image quality. For one of the way of image quality loss is imageblur, through the degree of image blur can evaluate the image quality. Edge is one of the mostcrucial features of the image. When a Gaussian blur occurs in the image, the edge width isincreasing, which is main features of the image. The width can be used to evaluate the imagequality.Aiming at the imprecise of blur measure for the inaccurate locating problem, this paperproposes a method that uses a multi-scale to obtain edge width. The follow is the maincontent of this paper.(1) First is establishing one-dimensional edge model of Gaussian blurry. Whilesearching the Gaussian blur, this paper set up a one-dimensional Gaussian blurred edge andthe first derivative of the Gaussian curve model. It put forward the width of edge represent thedegree of Gaussian blurs.(2) It discusses the measure of one-dimensional Gaussian blurry. The one-dimensionalfirst derivative model of the edge as the object, it structure multi-scale Gauss-Laplace filterbank. The maximum compensation responses of filtering reconstructed one-dimensional edgeobject and obtain the corresponding filter scale at this time. It is successful measure the blurof one-dimensional Gaussian blurred edge using reconstruction corresponding to maximal ofone-dimensional scaling edge width.(3) It researches the algorithm of Gaussian blurred measure of images. The edge’sgradient magnitude of the edge Gaussian blurred image make for a gradient line, with crosssection of the Gaussian function and direction, so construct multi-scale Gauss-Laplace filterbank with multi-angle. In the gradient direction, the maximum compensation response of themulti filters for the gradient line reconstructed energy image s and scale image of gradient line.The measure of Gaussian blurred image indicated by the average width of edge, whichstatistics the width in the scale image and centerline, which get form the rebuilding line usingMaxima.(4) It proposed the evaluation of measures method. The stability of the methoddetermined by the smallest linear deviation squared of Least-squares method for blurrymeasure values. The curve is formed by the measure value and the actual blur degree of thesame scene image. The first is calculating the average slope of the curve. The second is normalized average gradient. Last is statistics the variance, which determines the stability ofmethod. The quality of method is assessed by two indicators measure, the minimum deviationand average slope’s variance.In this paper, the width of edge is adaptive detection along with reduce the impact ofhuman factors. The result of comparative experiment show that the proposed methodimproves the accuracy of blur measures, what more, the stability and robustness are better forthe different image.