Speaker
Lucy Reynolds
Description
The two current main methods to measure shear, model-fitting and moments-based, both suffer from noise bias for galaxies with lower SNR. As Euclid has very strict requirements for shear bias, there is motivation to remove noise from images before performing the shape measurement to increase the precision. We present two deep learning algorithms: one to denoise galaxy images and another to measure the shapes of galaxies, with results comparable to GalSim adaptive moments and 8.2 times greater performance speed. Currently, we are developing a joint training framework to simultaneously train the denoiser and shape measurement in order to optimise the denoising for the shape measurement.
Primary author
Lucy Reynolds
