Image Representation Via a Finite Radon Transform
IEEE Transactions on Pattern Analysis and Machine Intelligence
A Fast Number Theoretic Finite Radon Transform
DICTA '09 Proceedings of the 2009 Digital Image Computing: Techniques and Applications
An exact, non-iterative Mojette inversion technique utilising ghosts
DGCI'08 Proceedings of the 14th IAPR international conference on Discrete geometry for computer imagery
On Constructing Minimal Ghosts
DICTA '10 Proceedings of the 2010 International Conference on Digital Image Computing: Techniques and Applications
Exact, scaled image rotations in finite Radon transform space
Pattern Recognition Letters
DGCI'11 Proceedings of the 16th IAPR international conference on Discrete geometry for computer imagery
Construction of switching components
DGCI'06 Proceedings of the 13th international conference on Discrete Geometry for Computer Imagery
DGCI'11 Proceedings of the 16th IAPR international conference on Discrete geometry for computer imagery
Hi-index | 0.00 |
Ghost images contain specially aligned pixels with intensities that are designed to sum to zero when projected at any of a preselected set of discrete angles. Ghost images find use in synthesizing the content of missing rows of image or projection space from data that contains some deliberate level of information redundancy. Here we examine the properties of ghost images that are constructed through a process of iterated convolution. An initial ghost is propagated by cumulative displacements into other discrete directions to expand the range of angles that have zero-sum projections. The discrete projection scheme used here is the finite Radon transform (FRT). We examine these accumulating ghosts to quantify the growth of their dynamic range of their pixel values and the spread of their spatial extent. After N propagations, a pair of points with intensity ±1 can replicate to produce a maximum total intensity of 2N. For the discrete projections of the FRT, we show that columnoriented iterations better suppress the range and rate of growth of ghost image values. After N row-based iterations, the peak pixel values of FRT ghost images grow approximately as 20.8N. After N column-based iterations, the peak pixel values of FRT ghost images grow approximately as 20.7N. The slower rate of expansion of pixel values for column iteration comes at the expense of fragmenting the compactness of the set of FRT projection angles that are chosen to sum to zero.