Denoising And Deblurring Poissonian Images Based On Variational Model

Digital image is often degraded during the process of acquisition,transmission.Degradation of image will not only damage to visual effects seriously,but also bring great difficulties for the subsequent pattern recognition and high-level vision analysis.Image restoration is a classical problem in image processing.While in the imaging of astronomy,biomedical and other fields,images are often corrupted by Poisson noise and the Poisson noise is not the common sense of the additive or multiplicative noise.So additive noise or multiplicative noise image restoration methods can not be directly used in Poisson noise.Poisson image restoration has become one of the hot spotts in the current research in the field of image restoration.In this paper,we study denoising and deblurring for the image corrupted by Poisson noise with the foundation on variational and regularization framework.Then,we explore the biomedical image segmentation problem under the foundation of poisson image restoration.The main contributions are as follows:A nonlocal total variation based image Poisson denoising model is proposed.In the model,statistics properties of Poisson noise and variance stable are considered,the similarity Gaussion weight of noisy image patch are replaced by nonlocal total variation.Variable splitting technique and alternating iterative method are used to solve the new model.Numerical results demonstrate that our method can preserve edges and textures effectively while denoising.A two-step Poisson denoising method is proposed for image obtained in the low photon counts.First,we enhance useful information by inpainting the image.Then a denoising step is performed by structure-preserved denoising method.For image inpainting model,Euler's elastica energy is used as a prior regularization term.And the Alternating Direction Method of Multipliers(ADMM)method is applied to solve the optimization problem.While for image denoising,the nonlocal total variation regularization is used for preserving the structure information.Experiments show the validity and efficiency of our proposed method for images with high or low Poisson noise.For the blurred image with Poisson noise,we first divide the image into cartoon and texture component.For cartoon component,fractional-order total variation regularization is used to eliminate staircase effect.For texture component,non-local total variation regularization is applied to preserve more details.Moreover,the background of the texture component is almost piecewise constant,which is great benefit to eliminate the artifacts resulted from the non-local regularization.We develop an alternating iterative algorithm to solve the proposed multi-term regularization model.Experiments show that our method performs well both in objective criterion and visual fidelity.After pre-processing for Poisson image restoration,Poisson image segmentation is researched further.From the perspective of human visual,we assume that pixels of the same subclass and manually labeled samples are very similar.Combined with the distribution of image space,an interactive segmentation method based on similarity search and spatial constraint is proposed.Firstly,features of manually labeled sample point are extracted.Then,a new image segmentation model is established by combining the similar fitting with spatial priori image.Finally,by using the continuous maximum flow method to solve the model,we can segment the image fast and effectively.The experimental results show that the pre-processed image achieve better segmentation outcomes due to the poisson denoising methods proposed,which have a better performance for detail-preserving.