Unsupervised Deep Image Stacking
Supervisor
Suitable for
Abstract
Description: Image stacking is a commonly used technique in astrophotography and other areas to improve the signal-to-noise ratio of images. The process works by first aligning a large number of short exposure images and then averaging them which reduces the noise variance of individual pixels. In this project you will investigate the use of neural networks for performing the stacking process in an “unsupervised” manner. This can be accomplished by predicting a distortion field for each image and using a consistency objective that tries to maximise the coherence between the undistorted images in the stack and the final output. During the project you will evaluate this approach and compare it to traditional image stackingPrerequisites: Suitable for those who have done the Machine Learning course. Having also done Computer Graphics would be useful.