ANCHORED DEEP NETWORK FOR ENHANCED SINGLE-IMAGE SUPER-RESOLUTION IN DEFENCE APPLICATIONS

Abstract

It is a difficult issue in clever monitoring apps to analyze images and videos in real time. As a result of network
constraints, many apps must make sacrifices between frame rate and sharpness. As a result, super-resolution
imaging has become a standard feature of many security systems. The Using picture previous to its maximum
potential has been shown to boost the efficacy of current image super-resolution algorithms. However, earlier
images are rarely considered by existing deep learning-based picture super-resolution techniques. Therefore, one of
the open questions for deep-network-based single-image super-resolution techniques is how to make optimal use of
image previous. In this article, we use transfer learning to ensure that our suggested deep network accounts for the
image previous, thereby bridging the gap between the conventional sparse-representation-based single-image
super-resolution techniques and the deep-learning-based ones. There is still the issue of how to prevent neurons
from compromising on various picture elements when using a deep learning-based single-image super-resolution
technique. In this work, the picture patches are fixed to the lexicon atoms so that they can be sorted into classes.
Because each neuron is trained on regions of the picture with comparable clarity, the network is better able to
retrieve high-frequency information.