Signal Processing - September 2016 - 137

(a)

(b)

(c)

(d)

Figure 4. Elemental maps, obtained on Vincent van Gogh's Patch of Grass, showing the hidden portrait of a woman. (a) and (b) show the Sb distribu-

tion, while (c) and (d) show the Hg distribution. (a) and (c) were acquired with macro-XRF at a synchrotron source, while (b) and (d) are results of in situ
measurements by means of Instrument B. (a) and (c) were acquired with a step size of 0.5 mm and 2 seconds dwell time in two days, while (b) and (d)
were acquired with a step size of 1 mm and a dwell time of 5.1 seconds in six days. (Images and figure caption used with permission from [28].)

of electron density, the method requires microcrystallinity
in the analyte. Thus, the technique cannot be used on
amorphous materials or materials that do not diffract
X-rays well, and in this regard it is inferior to XRF. However, because the diffraction pattern of a crystalline substance
is essentially a fingerprint, it provides direct chemical
information about the analyte, and in that regard it is superior to XRF. Because XRD typically requires greater photon flux than XRF, the method was more resistant to
migration from synchrotrons to transportable scanning
methodology. Fortunately, those problems are being solved
[5]. In addition to providing positive chemical identification
of materials present, XRD also offers the advantage that,
because it requires higher energy X-rays, it provides greater
depth penetration. Combined with XRF macro scanning
data cubes and hyperspectral imaging cubes from the UV,
vis, and IR, these techniques, operating synergistically,
allow unprecedented insights into the composition of cultural heritage objects, with all of the attendant implications
for art history and art conservation.

Conclusions
In this article, we surveyed how computational imaging has
impacted five key areas of cultural heritage science. There are
three key features that have resulted in these techniques making a significant impact on the cultural heritage community.
The first is the proliferation in recent years of image sensing
technology, which has spawned technological advances in
new imaging modalities such as XRF, XRD, hyperspectral,
etc. The second feature is that recent advances in these new
imaging modalities has given accessibility to entirely new
types of information latent within the artworks held by museums. The third feature is the ability to visualize information
about artifacts intuitively in the form of images, which has
made this information much more accessible and comprehensible to nonexperts.

Computational imaging of cultural heritage is opening up
many new avenues for investigating the technical art history of
objects and to assess the condition of works of art that will aid
in their long-term preservation. There are several areas of computational imaging that have not been thoroughly explored on
cultural heritage objects. Also compressive sensing and sparse
imaging could significantly improve sensitivities especially for
conditions where low light is necessary for light-sensitive materials and when increased imaging speeds are necessary for experiments that cannot be conducted in the public spaces of museums
over days (as in macro X-ray scanning). Improved material databases with bidirectional reflectance distribution function data
[30] could lead to advances in reconstruction algorithms that
produce more accurate image archives and renderings. Scalability is another principal obstacle. For example, a comprehensive
measurement of the chemical composition and spatial structure
of layers of paint in an entire work of art could provide new and
valuable tools for art historians and conservators.
Another important direction that has not been covered in this
article is the dissemination, visualization, and display of the
great body of visual information now being captured by museums and galleries around the world. For instance, augmented
reality is projected to strongly impact the museum visitor's
experience in coming decades. Finally, for the computational
imaging field, it is important to note that artworks provide
fantastic test scenes that can inspire researchers to push the
envelope by providing new imaging and display techniques
that can probe the complex light-material interactions inherent in so many works of art. In this regard, it is the hope that
cultural heritage can serve as a catalyst for novel research in
computational imaging.

Authors
Xiang Huang (xianghuang@gmail.com) is a postdoctoral
researcher in the Mathematics and Computer Science Division
of Argonne National Laboratory. His interestes are solving

IEEE SIgnal ProcESSIng MagazInE

September 2016

137

Table of Contents for the Digital Edition of Signal Processing - September 2016