Research On Color Image Zooming Based On Variational Regularization

Image zooming is one of the main research problems in image processing.Its main purpose is to change the size of the original image to meet certain specific requirements,such as magnify the medical image so that the doctor can diagnose and treat;scale the same photo at different scales to meet the requirements of displaying the same image on machines with different resolutions.With the gradual improvement of people's requirements for the quality of life,color images gradually come into the field of vision and take the place of gray images,so it is very important to acquire clear color images in a timely and effective manner.This paper mainly studies the problem of color image zooming,proposes several image zooming models based on variational regularization,and solves the proposed model.The main research contents and innovations are as follows:1. An image zooming model based on the second-order Total Generalized Variation regularization?TGV?regularization is proposed multi-channel color image zooming and we specifically design a more effective algorithm based on the Alternating Direction Method of Multipliers algorithm?ADMM?rather than the prime-dual algorithm to solve the model.In RGB color space,we amplify each color channel separately to enlarge the color image.Numerical results show that compared with the prime-dual algorithm,the ADMM algorithm achieves better amplification effect both in visual effect and quantitative comparison.2. A new image zooming model based on Nonconvex and Nonsmooth Total Generalized Variation?NTGV?is proposed to save the edge information of the image effectively as the NTGV regularization combines the advantages of nonconvexity and TGV regularization.We adopt the Iteratively Reweighted ₁l Algorithm?IRLA?to solve the proposed nonconvex model.Simulation experiments show that the model in this paper is superior to the scaled-up model of fractional partial differential equations in both visual effect and quantitative comparison.3. A new image zooming model based on Overlapping Group Sparsity?OGS?and high order non-convex TV regularization is proposed.The ADMM algorithm and Majorization-Minimization algorithm?MM?are used to solve the proposed model.The simulation experiment shows that compared with the TGV model,this model can better protect the image edge information and has a better numerical performance.