Research On Super Resolution Of Multi-Frame Images Under Large-Scale Displacement

High-resolution images contain richer details,higher definition,and they play an important role in actual production and living applications.However,in the image acquisition process,the images often unable to meet the resolution requirement in practical applications since imaging process may be influenced by defocusing,jitter,and noise,etc.The image super resolution reconstruction technology can increase the spatial resolution of the imaging without changing the optical systems through the computational imaging theory.It has the characteristics of flexibility,low cost,and strong adaptability.The super resolution algorithm based on Variational Bayesian can reconstruct high-resolution images with multi-frame images.This algorithm uses probability distribution estimation instead of point estimation to reduce the propagation error introduced by multi-parameter estimation and to improve the reconstruction accuracy.However,due to the simple imaging model and low precision of motion estimation,the effect of the current Variational Bayesian algorithm under large-scale displacement conditions is not ideal.In this paper,the large-scale displacements and angle changes during shooting are considered,a dual-level motion estimation algorithm under large-scale displacement is proposed,which can greatly improve the image reconstruction effect.The main research work in this article includes the following sections.(1)The procedure of imaging link is improved.The projection transformation generated by the lens shift during the shooting process is added to the imaging link based on the in-depth study of the Variational Bayesian super resolution reconstruction algorithm.When correcting the deformation,the scale-invariant feature transformation algorithm is first used to find and match the image feature points.Then the random sample consensus algorithm is designed to eliminate the mismatched points and estimate the optimal solution of the transformation matrix.Finally,the estimation errors can be eliminated through the projection transformation of images.(2)The motion estimation process is optimized.A dual-level motion estimation algorithm is proposed to improve the motion estimation accuracy under the condition of large-scale displacements.The algorithm combines the motion estimation of whole and sub pixels.First of all,the projection matrix is used to get the rough motion estimation of the whole pixel and the image is adjusted.Then the affine transformation is combined with the Bayesian framework to accurately the sub-pixel motion estimation,greatly improves the displacement limit of Variational Bayesian algorithm.(3)Multi-frame super resolution reconstruction under the large-scale displacement condition is proposed.Based on the Variational Bayesian algorithm,the reconstruction process is optimized.One target image is segmented according to the overlapping of multi-frame images.The interpolation algorithm is applied for non-overlapped regions to reduce the computational complexity,while the super resolution reconstruction algorithm is implemented for overlapped regions,which ensures full use of non-redundant information among sequence images.(4)The experiments of super resolution reconstruction under large-scale displacement are carried out.Several sets of displacement conditions are designed and the effects of reconstructed images are compared and analyzed.The experimental results show that in the case of large-scale displacement,the reconstruction results of this proposed algorithm are superior in both visual effects and data indicators,verifying its effectiveness and practicality.