This repository contains files for computational astrophotography deblurring.
See the written report at Computation_Star_Tracking_Paper.pdf
This file executes the full deconvolution pipeline on an unblurred star image. It simulates a blurring effect around a given center location in the image, then maps the image to polar space and performs deconvolution with a linear blur kernel, then maps the image back to cartesian coordinates and applys brightness and contrast enhancement. The script has options for choosing the rotation center, deconvolution method, number of iterations, hyperparameters, regularization constants, and enhancement parameters.
This file tests different methods for identifying the center of rotation of a star trail image. It contains options for least-squares fit, convex optimization implementation, and Hough transform method.
This file runs the full deblurring pipeline on a real star trail image. It uses the Hough transform to obtain the center of rotation then deconvolves the blur in the polar domain using ADMM with TV prior.
These folders contain supplementary functions and scripts.
This folder contains different functions used to match circles to images or image coordinates for the purposen of determining star trail center of rotation.
This folder contains functions for different deconvolution methods.
This folder contains miscellaneous utlity functions, such as mapping polar to/from cartesian, padding and unpadding the image, cropping away artifacts, normalization, contrast enhancement.
This folder contains miscellaneous scripts used during testing.
This folder contains sample images from stages in the image processing pipeline.
This folder contains various figures and plots of results generated.
This folder contains stock photos obtained from the internet of star scenes and star trail images.
This folder contains the images used to evaluate the circle detection methods (least-squares, hough transform, and CNN).
This folder contains all the files for training and running the CNN
This file creates a dataset of images with labels of rotation centers of simulated star trail data.
This file trains a CNN with architecture in the build_model() function on the training data in the folder neural_network_data. This function can also be used to evaluate the trained network on a given image.