Image Algorithm Summary

These tables list the image reconstruction algorithms available in the RHESSI software grouped by whether they are based on visibilities or not. Click the Name of the algorithm for more detail on that method.

Calibrated Eventlist-based Algorithms
Algorithm Description Advantages Disadvantages
Opens internal link in current windowBack Projection Most basic method of image reconstruction. Multiplies the calibrated eventlist flux by the collimator modulation patterns and sums the result to construct a 'dirty map' of the image. Very fast. Linear. Simple. Can be useful in finding the flux centroid with very low signal levels. Poor quality images with sidelobes. No reliable source sizes.
Opens internal link in current windowClean Sequentially identifies peaks in the residual back projection map and removes contributions from the associated sources convolved with the Point Spread Function. The reconstructed image is the sum of Gaussians based on the point source fluxes. Relatively fast. Positive fluxes. Removes sidelobes. Prone to photometric errors. Likely to misinterpret extended sources.
Opens internal link in current windowForward Fit Assumes the source(s) can be presented as a simple functional form with a few free parameters (for example, a circular Gaussian source) and looks for the parameters that produce a map that is consistent with the data. Fast. Positive fluxes. Provides source sizes and errors on parameters. Requires that assumptions concerning the number of sources and their shapes are accurate. Does not adequately search parameter space.
Opens internal link in current windowPixon Attempts to reconstruct the source through a sequence of maximum entropy processes while constructing a spatially local regularization. The regularization is done by smoothing the image according to the current Pixon size scale The goal is to construct the image with the fewest degrees of freedom (the fewest pixons) that is consistent with the observations. Robust. Accurate image photometry. Best method to image extended sources in the presence of compact sources. At more than 65x65 pixels the process time increases rapidly. On 2019 systems images should be made in under a minute. Pixon_sensitivity is an additional regularization parameter that needs to be increased for >1e5 counts. Should not exceed 2.
Opens internal link in current windowEM An iterative reconstruction method based on the Lucy-Richardson Maximum Likelihood method. At each step, uses a back-projection of the ratio of observed counts to the expected counts to modify current image. Does not use the FLATFIELD back-projection and modifies image by multiplying by a always positive image therefore rigorously maintains positivity in the image. Speed is comparable to Pixon, sometimes faster. Similar photometric qualities to Pixon. Fairly robust to choice of detectors. Does not explore solution space as thoroughly as Pixon. Images will break up if pushed too hard. Avoid using EM with very high # of counts (>1e6) as there is no allowance for systematic errors in the counts.
Opens internal link in current windowMEM Sato

(no longer supported)

Maximum Entropy Methods (MEM) algorithm finds the 'most plausible' solution which has the maximum configuration entropy among those solutions that are consistent with the observation. Relatively fast. Positive fluxes. Unsupported.


Visibility-based Algorithms
Algorithm Description Advantages Disadvantages
Opens internal link in current windowMEM NJIT Maximum Entropy Method algorithm based on visibilities. This is a 2-D version of radio astronomy's SSMEM provided by NJIT. Fast. Accurate image photometry and morphology. Capable of over-resolution. Subject to source breakup particularly when using data from the finest grids.
Opens internal link in current windowMEM GE Maximum Entropy Method algorithm based on visibilities provided by Univ. of Genoa. Fast. Accurate image photometry and morphology. Not as prone to source breakup as MEM NJIT because of greater control of the regularization. ...
Opens internal link in current windowVIS FWDFIT Forward-fit algorithm based on visibilities. Like the Forward-Fit algorithm, assumes there are a limited number of parameterized simple individual sources. Adjusts the source parameters such that the model-predicted visibilities agree best with the measured visibilities. Fast. Provides uncertainties on source parameters. Requires that assumptions concerning the number of sources and their shapes are accurate. Results are not always stable.
Opens internal link in current windowUV SMOOTH Performs a Fast Fourier Transform of 2-D spline-interpolated visibilities. Fast. Good photometry. Vulnerable to missing visibilities. The interpolation can break down when the FOV to grid scale size is too large.
Opens internal link in current windowVIS CS Compressed Sensing image reconstruction. Attempts to make the reconstruction by matching the visibilities from a minimum set of shapes pre-determined by a dictionary of shapes. Fast. Accurate photometry and morphology. Robust. ...
Opens internal link in current windowVIS WV Finite Isotropic waVElet transform Compressed Sensing (also known as 5-CS or FIVE-CS) Notable spatial accuracy. Significantly reduces the ringing effects due to the instrument point spread function. ...
Opens internal link in current windowMEM Vis

(no longer supported)

Similar to MEM Sato, except works with visibilities instead of counts. Relatively fast. Positive fluxes.

Unsupported.