Image Super-Resolution
Deep Internal Learning vs External Learning
Super-Resolution Explained
Super-resolution (SR) is the process of recovering high-resolution images from low-resolution inputs, mostly by understanding the way that human vision works. Normally when you look at an image captured with a low resolution camera (such as a smart phone camera), it will look grainy and pixelated. However, if you zoom in closer and focus on an area of the image, your brain fills in the missing details.
SR systems take low-resolution image(s) as an input and produces high-resolution version (increased the number of pixels) of those provided image(s) after doing some kind of processing. These systems use a variety of SR techniques which could be single image based or multiple image based but in this work we will be focusing only on single image techniques based on deep learning.
Figure 1: Resolution of this image has been increased to 960x640 from 480x320.
How Super-Resolution helps
Today, visual data, be it images or videos, is the most widely captured form of data. It has many applications in our lives ranging from recording memories from our daily lives to saving data from space satellites. The high-quality images reconstructed by using the approaches in [1] and [3] can dramatically improve the visual quality for all of such applications e.g. surveillance, archiving and medical image analysis etc. Super resolution is a high-quality image synthesis technique which is applicable to various images, including astronomical, geographical, and microscopy images. The increasing resolution of the data produced by satellites, medical procedures, geographical studies etc. is of critical importance not just to the scientific community but also to numerous other sectors, ranging from the sectors of civil protection to urban planning or energy and many more.
​
​
​
​
​
Figure 2: SR in medical imaging and satellite imaging.
Some other applications include: Video information enhancement, Surveillance, Biometric information identification etc.​​
​
​
​
​
​
Figure 3: SR Applications.