Deblurring Analysis Of The Blurred Image And Estimation Of The Blur Kernel Based On The Hand-held Facilities

When people acquire image by the hand-held shooting facilities in daily life,motion blur often happens, resulting from the motion between the facilities and theobject. Accordingly, the restoration of motion blurred image aims at extractingeffective information from blurred image and establishing the blur model in order torestore the original clear image as much as possible. This article consists of theresearches about initializing the blur kernel related to motion information andrestoring the image and so on. Among these researches, the estimation of initializingthe blur kernel is the key point in the process of deblurring, and also is the coreanalysis of this article.At first, the blur model is established and the principle of blur kernel generation isanalyzed in two cases detailedly. The first case is that the blur kernel of images isestimated through the gradients of boundaries in flat background. Through analyzingthe gradients of boundaries in the blurred images, the motion features are extractedwhich lead to blur. Then the blur kernel initialization model is built. This experimentshows that the effective initialization blur kernel is extracted, and the dependence ofrecursion optimization in the process of deblurring is weakened. The time costreduced and the effects are good, especially for images in flat background and theones with significant changes near boundaries. The second case is the blur imagesanalysis of the hand-held facilities within inertial measurement sensors when they areshaken slightly. Through building the related motion blur model, this article analyzesthe data of the inertial measurement sensors when the hand-held shooting facilities areshaken slightly in acquiring an image. Then the initialization of the blur kernel ispresented, and finally the theoretic initialization model about the complex planarmovement is proposed. This experiment shows that this model can initialize the blurkernel efficiently and effectively.Then, by the wiener filtering method and Richardson-Lucy iterative algorithm,this paper discusses the image restoration, aiming to achieve the same restorationeffect with the simplest method. The experiments show that with the greatercorrelation between the blur kernel and the motion blur image, the effect of imagerestoration will be greater and the dependence recursion in the process of deblurring will be less.