A Novel End-To-End Network for Reconstruction of Non-Regularly Sampled Image Data Using Locally Fully Connected Layers

Symbolic picture for the article. The link opens the image in a large view.

In this paper, we propose a novel end-to-end neural network to reconstruct high resolution images from non-regularly sampled sensor data. The network is a concatenation of a locally fully connected reconstruction network (LFCR) and a standard VDSR network. Altogether, using a three-quarter sampling sensor with our novel neural network layout, the image quality in terms of PSNR for the Urban100 dataset can be increased by 2.96 dB compared to the state-of-the-art approach. Compared to a low-resolution sensor with VDSR, a gain of 1.11 dB is achieved.

Further information about the novel consistency checks is provided in the following presentation.


This contribution was presented at MMSP 2021 (6.-8. October 2021).

Research at the chair is regularly being published at international conferences and impactful journals. In times of virtual conferences, many organizers have resorted to presentations in the form of pre-recorded videos that can be accessed on-demand by attendees. We use this opportunity to make our research available to a broader public by sharing some of our contributions on our website.