PROJECT TITLE :

LCSCNet Linear Compressing-Based Skip-Connecting Network for Image Super-Resolution

ABSTRACT:

For image super-resolution, we present in this research a short but effective network design dubbed the linear compressing based skip-connector network (LCSCNet). A linear compression layer is created in LCSCNet for skip connections, which joins previous feature maps and differentiates them from newly investigated feature maps, in comparison to two sample network architectures with skip connections, ResNet and DenseNet. In this way, the suggested LCSCNet has the advantages of DenseNet's differentiate feature treatment and ResNet's parameter-economic form. As a result, we also present an adaptive element-wise fusion technique with multi-supervised training in order to better use hierarchical information from both low and high levels of diverse receptive fields of deep models. The usefulness of LCSCNet has been demonstrated experimentally and in comparison with leading-edge algorithms.