Anti-noise Spatial Super-resolution Of Light Field Epipolar Plane Images

Visual information is significant important to mankind.Digital image is one of the most extensive media of visual information.To capture three-dimension information of a target scene,the light field imaging can simultaneously capture the spatial and angular information of light rays by using a pair of two-paralleled-plane model.Comparing with traditional imaging,the light field spatial resolution is still limited by optics design and manufacture,which prevent the further development of light field applications.In recent years,various different methods have been proposed to address the light field super-resolution reconstruction.However,the light field imaging result could be significantly deteriorated by the low luminous flux and sensor noise,which are more severe than traditional imaging.Traditional super-resolution reconstruction can be vulnerable to deal with such strong noise.Therefore,we explore the anti-noise light field super-resolution reconstruction in this paper.This work can be summarized as follow:(1)Based on the theory of image local denoising,this paper proposes to apply denoising in horizontal and vertical directions receptively by introducing a local similarity of epipolar plane image(EPI).Considering the angular continuous of the light field,this paper fused the horizontal and vertical denoised results using a the contrast pyramid and the directional filter based algorithm,which preserves the linear structure of EPIs.(2)Leading by the projection onto-convex sets(POCS)super-resolution,we introduce an improved light field POCS super-resolution algorithm.To take the advantages of angular information,we propose to super-resolve the central view by re-projecting four-neighbor images.To ensure the anti-noise performance,we proposes the bi-pass denoising algorithm to further reduce the "ringing" artifacts.(3)Further,a visual operation framework of the proposed methods is conducted to facilitate the practical use of light field de-noising and super-resolution.It enables both the integrated light field datasets and external user-defined datasets.In the preprocessing stage,The system provides five types of noises with different noise strength.A variety of different algorithms are also provided to test the performance of super-resolution reconstruction.In summary,this paper aims at the problem of light field image super-resolution under strong noises conditions.This study explores the spatial super-resolution and denoise in light field EPI space.On multiple light field datasets,the experiment results show that the effectiveness and robustness of the super-resolution in presence of strong noises.