The problem with the good contrasts
The two richest X-ray contrasts are the two hardest to photograph.
Fluorescence tells you which element sits where, at trace concentration, with a specificity no absorption measurement approaches. But a fluorescence photon leaves the sample in a random direction and carries no information about where it came from. The usual answer is to raster-scan a focused beam: exquisite maps, one pixel at a time, and a full tomogram that takes days of beamtime nobody has.
Small-angle scattering tells you about structure far below the resolution of the detector. Measuring it properly means a point-scanning SAXS instrument, and a full tensor tomogram — the orientation of that sub-pixel structure in three dimensions — is a heroic experiment.
Structure the illumination, not the detection
Single-pixel imaging inverts the problem. Rather than asking the detector to resolve position, we pattern the beam and record a sequence of total intensities on a detector with no spatial resolution whatsoever. Each measurement is one projection of the sample onto one known pattern; the image is recovered by solving the resulting inverse problem. Choose the patterns well — and choose the regulariser well — and far fewer measurements are needed than the number of pixels you recover.
This is the same intellectual move as everything else in the group. In modulation-based imaging a mask encodes the wavefront so a camera can read phase; here a patterned beam encodes position so a bucket detector can read chemistry. In both cases the physics is put in the forward model and the pixels are recovered by inversion.
What we are building
XRF ghost tomography — element-specific three-dimensional maps from structured illumination and a single-element fluorescence detector, replacing the raster scan with a compressive acquisition.
SAXS tensor tomography — full reconstruction of the orientation of sub-pixel structure in a volume, from projections that mix scattering directions together.
Both run under the MIAI FANFARES chair, held jointly with Nicolas Ducros at CREATIS — whose group built much of the single-pixel imaging methodology — and Nicola Viganò at CEA IRIG. Two doctoral contracts run on these questions until 2029.
To add: a figure of the pattern sequence and a reconstructed elemental map, and a sentence on which beamlines this runs at.