Research On Blur Kernel Estimation And Deblurring Of A Single Image

Image deblurring aims to generate a high quality sharp image from its blurred version.Image blur widely exists in the process of imaging because of a variety of factors,including camera shake,defocus,moving objects and the imaging quality of sensors.Image blur directly affects the quality of images,and causes losses in the image information.Hence,restoring a sharp image from its blurred version is important in image processing field.In this paper,we find that the existing image deblurring methods can eliminate the global motion blur which is caused by camera shake.However,these deblurring methods do not apply to the image degraded by spatially-varying defocus blur or partial motion blur.In order to eliminate these two types of blurs more effectively and further extend the scope of application of image deblurring technique,we start addressing the problem of deblurring images degraded by defocus blur and partial blur based on existing image deblurring algorithms.The main works and contributions are introduced as follows:(1)We come up with an effective image deblurring method based on multi-scale defocus cues by combining defocus blur estimation method and image deblurring technique.Firstly,we present a defocus map generation algorithm by using the multi-scale strategy,which includes three steps: defocus blur estimation,sparse defocus map refinement and sparse defocus map interpolation.Secondly,we present an efficient multi-parameter regularization model which can deblur images with a spatially-uniform kernel.Finally,we use the full defocus map to restore a latent all-in-focus image from the original blurred version.The experimental results demonstrate that our method obtains defocus maps with fairly high accuracy and restores high quality all-in-focus images.(2)We combine the partial blur classification method and image deblurring technique together and present an image deblurring method based on deep learning to eliminate the partial blur.Firstly,based on the existing partial blur classification methods,we present a deep learning framework with a stacked auto-encoder,which can be used to detect and classify blurred regions in a partially blurred image,as well as to mark the blurred regions to eliminate the partial blur effectively.Then,we treat these blurred regions as mask layer,and only remove the motion blur in these blurred regions.In this way,we can not only eliminate the partial blur but also reduce image distortion,and finally restore a satisfactory deblurred image.