| Motion blur is a common image degradation phenomenon,which is caused by the relative movement of the subject and the camera due to camera shaking or object motion during camera exposure.In recent years,a large number of state-of-the-art deblurring algorithms have been proposed,but most of them are based on the assumption of linear blur model and slight noise.However,under extreme shooting conditions,blurred images usually contain considerable amount of outliers(such as non-Gaussian noise and saturated pixels)that do not meet the linear model.Typical scenario,such as shooting at night with a hand-held device.Due to the low light condition the camera needs longer exposure time,it is easy to introduce more noise during the imaging process,and inevitably shake will cause motion blur.At the same time,various artificial light sources and specular reflections saturate parts of the image.These outliers violate the linear blur assumptions and cause severe artefacts in the restored image.In order to deal with the interference of different types of outliers,the main achievements of this paper are as follows:1.An outlier detection and deblurring model based on joint sparse and maximum entropy prior is proposed.By iteratively performing outlier detection and image deblurring,the algorithm can effectively detect and suppress sparse outliers(such as impulse noise)and finally restore a clear image.Compared with the classical denoising algorithm,the algorithm in this paper has better adaptability to different density impact noise.2.Unlike impulse noise,the distribution of saturated pixels is clustered.By analyzing the influence of the highly saturated region on the deblurring algorithm,an algorithm correcting ringing effect around the high saturation region is designed as a post-processing step of the restored image.Experiments on partially saturated blurred images show that the SSIM of the final restored image is improved by about 0.1 on average compared with other state-of-the-art algorithms. |
PDF Full Text Request