Quality Assessment And Parameter Estimation For Blur Image

Blur is caused by the relative motion between the camera and the target while images are collected.This paper mainly focuses on image quality evaluation algorithms for defocus blur,global and local linear motion blur.Besides,blur parameter estimation methods for global and local linear motion blur are also discussed.The validity of the proposed methods is verified by experiments.The main contributions of this paper are as follows:1.A defocus blur image quality assessment algorithm based on large scale structure and salient edges is proposed.In this method,the weighted least square filter is used to extract the large scale structure of the image,so as to eliminate the interference of dense textures with small edge width.Then the proportion of salient edges with different widths is calculated to eliminate the interference of image content and contrast.Furthermore,the mean value of the maximum part of gradients is calculated.Finally,the support vector regression approach is used to obtain a reasonable relationship model between these features and image quality.Experiments demonstrates that images scores gained by this method have good consistency with the subjective evaluation values.2.A local linear motion blur image quality assessment method based on the directional gradient auto-correlation function is proposed.According to the statistical characteristics of high frequency and low frequency,it is difficult to distinguish blur regions and smooth background.The method analyzes the characteristics of the gradient auto-correlation function in the movement direction and uses it to extract the motion blurred target and evaluate image quality.This method has the ability to extract pedestrians,bicycles and cars on the road.3.A local motion blur parameter estimation method based on the change of pixel value in overlapping regions of target and background is proposed.Dark stripes in the frequency domain of motion blur region are not obvious.The method analyzes the characteristic of pixel change in overlapping regions.Firstly,the pixel distribution is estimated when the edge is sharp.Secondly,blur kernels with different scale are used to fuzz sharp edges.Finally,we compare gradient values between them and the pixels in overlapping regions,and choose the scale with the minimum difference.This method has good estimation accuracy.