The proposed paper addresses the inverse problems using a novel deep convolutional neural network (CNN). Over the years, regularized iterative algorithms have been observed to be the standard approach to address this issue. Though these methodologies give an excellent output, they still impose challenges such as difficulty of hyper parameter selection, increasing computational cost for adjoint operators and forward operators. It has been observed that when the normal operator of the forward model is seen to be a convolution, unrolled iterative methods take up the CNN form. In view of this observation we have proposed a methodology which uses CNN after direct inversion to find the solution for convolutional inverse problem. In the first step the physical model of the system is analyzed using direct inversion. However, this leads to artifacts which are then removed using a combination of residual learning and multi-resolution decomposition in CNN. The results show that the performance of the proposed work outperforms other algorithm and requires a maximum of 1 second to reconstruct an image of high definition.