Rendered Image Superresolution Reconstruction with Multichannel Feature Network

Abstract

In the process of film and television production, clear images can give the audience a real sensory experience, but high-resolution images require a massive amount of production time and highly specialized imaging equipment, which is not a cost-effective solution at the moment. To achieve a better cost efficiency during video production, we propose a multichannel featured superresolution network model that utilizes rendered low-resolution images according to their characteristics. This model includes a feature extraction layer, a series of subnetworks, and a reconstruction module. Inside the network model, a series of subnetworks are cascaded to improve the information flow from coarse to fine, which helps to fully extract the depth, normal vector, edge, and texture features from low-resolution rendered images to reconstruct the high-resolution image. Additionally, residual learning is introduced at each stage to further improve the reconstruction performance. We experiment with the model on the classic Disney Monte Carlo datasets and compare it with several related algorithms. The results show that our algorithm is able to reconstruct the image with clearer details and texture. Thus, our research not only helps to maintain the audience’s sensory experience but also increases the efficiency of film and television production, which also brings considerable economic benefits.

Research on Key Technologies of Super-Resolution Reconstruction of Medium and Long Wave Maritime Infrared Image

“…In order to ensure the rapid convergence of the network, the gradient clip is gradually set within a certain range [−τ, τ], where τ is set to 0.4 [25][26][27]. The activation function and regularization operation follow each convolution layer, and the negative slope of the activation layer is 0.2 [28][29][30]. We trained 90 generations of deep learning according to the above-mentioned set procedure, and the entire process take about 75 h. We observed that the loss function of the 85th generation network model was the smallest and chose this model for subsequent experiment.…”