Inverse problems are those in which information aboute a phenomenon is sought only from indirect measures. 
In the area of image processing, they include tomographic and deblurring methods, for example. When they are 
ill-posed, the discretization for computational solution of these problems results in ill-conditioned systems of equations, 
unstable in the presence of noise. One way to solve them is through the regularization method, treating them as optimization 
problems and restricting the solution space with a priori information of the respective solutions. In recent years, 
end-to-end deep learning solutions have been proposed to solve inverse problems. Instead of this approach, the present work 
studies the union of the paradigms of regularization and deep learning methods in the solution of ill-posed problems, 
investigating the development of new regularization terms from neural networks.