Study Of Fast Motion Image Deblurring Algorithm

Motion blur is one of the prime causes of poor image quality in digital imaging.It is caused by camera shaking or the movement of objects during the camera exposure process.Blind image deblurring is a classical image and signal processing problem,which aims to recover a blur kernel and a clear latent image from a blurry image.The main research contents of this paper include three aspects: study of blur kernel size estimation algorithm based on image cepstrum analysis and image spectrum analysis,study of mathematical model of motion deblurring based on salient region detection,study of adaptive iterative optimization for motion deblurring.Blur kernel size is essential to blur kernel estimation in image deblurring algorithm.Blur kernel size in earlier algorithms is an input parameter.In recent years,some algorithms can accurately estimate parameterized blur kernel.The none-parameterized blur kernel is more usual in the natural environment,while existing algorithms can't accurately deal with none-parameterized blur kernel.This paper first use image cepstrum to estimate both the size and shape of blur kernel.Then spectrum is used to estimate the blur kernel with small size.This method can accurately deblur most blurry images with small computational cost.Impressive progress has been made in estimating a complex motion blur kernel.Sparse priors(image gradient)and the multi-scale iterative framework contribute much to the success.In recent years,some researcher point out that image intensity is also sparse.While these existing algorithms based on image intensity is sensitive to the distribution of image intensity.In this paper,image intensity prior based on salient region detection is introduced.Similar with image gradient prior,salient intensity prior is also follow a heavy-tailed distribution.In the process of image deblur,the sparsity of salient value of latent image intensity will reduce.This method can accurately detect salient details with important information from blurry images with complex image intensity distribution.Because of these,image deblurring algorithm based on salient intensity prior can accurately estimate clear latent image.The number of iterations is essential to the estimation of latent image in the iteration process of image deblurring algorithm.In the multi-scale iterative framework used by earlier algorithms,the number of iterations in one scale is usually fixed in each scale.But the number of iterations depends on many factors,such as the complexity of blur kernel,the quality of initial kernel and the salient structures of blurry image.So a fixed number of iterations is not suitable for each scale.In this paper,an adaptive iteration strategy is proposed to adjust the number of iterations by evaluating the cepstrum of latent image intensity and the similarity of estimated kernel.This method can make the number of iterations is suitable for each scale,which is helpful to accurately deblur blurry image.Extensive experiments demonstrate that the proposed motion image deblurring algorithm can accurately deblur face images,text images,saturated images and dataset with complex background.Compared with existing image deblurring algorithms based on image intensity prior and image gradient prior,the proposed algorithm can accurately recovery clear image with small computational cost.