14 | New Scientist | 7 September 2019
News
“ENHANCE!” Artificial intelligence
is turning the TV crime drama
trope – in which a blurry image of
a suspect is made crisp and clear
by computer – into reality.
Vishal Patel at Johns Hopkins
University in Maryland and his
colleagues have produced an AI
that can automatically deblur
pictures of faces.
The system was developed in
collaboration with a researcher
at Adobe, a company that makes
software including photo-editing
programs. It could one day be used
to improve facial recognition on
long-distance surveillance images,
such as those taken from a drone,
says Patel. It could also be used as
a tool to automatically fix blur in
our everyday snaps.
Altering images in this way
is challenging because many
different types of blur exist
and it is difficult to determine
which kinds are present in order
to correct them. Examples
include motion blur, which
occurs when a photo is taken
while a camera is moving or
when a person is moving but
the camera is stationary.The team trained the AI by
providing it with thousands of
photos of people’s faces, both
clear and blurry. In each picture,
different parts of the face were
labelled, such as the eyes, nose
and mouth. Previously, AIs have
had difficulty deblurring these
kinds of features.
From this, the AI learned to
deblur features, including skin
and hair, and then combine theminto a final clear photograph.
The team tested the system’s
ability using 16,000 images that
had been blurred, comparing the
algorithm’s result with the
original flaw-free photo.
Its performance was assessed
using a structural similarity
index, where a score of 1 indicates
an exact match between two
images. The algorithm did
extremely well, scoring up to
0.96 (arxiv.org/abs/1907.13106).
Blur is just one of several factors
that can affect the quality of an
image, which is particularlyimportant if a photo is being used
for crime investigation purposes.
Others include different weather
and light conditions or a person
being partially obscured by objects
or other people, says Shufan Yang
at the University of Glasgow, UK.
High-speed winds and magnetic
fields can also interfere with
cameras, she says.
“The other thing that you
see in CCTV cameras is low
resolution,” says Patel. As a
result, the images they produce
are often small and lack detail.
To counter this problem,
software that creates extra
image detail has been developed
to increase resolution.
Many ethical questions need
to be answered before such
technologies can be used as
part of criminal investigations,
given the risk that an AI could
generate a clear picture of a
suspect that is different to
their actual appearance. When
combined with inaccurate or
racially biased facial recognition
algorithms, there is a possibility a
deblurred photo could lead to the
wrong person being identified. ❚“ Biologists have long been
seeking ways to correct
the mutations that stop
healthy sperm production”SPMEMORY/GETTYGeneticsCRISPR could
fix one form of
male infertilityMALE mice with a mutation that
stopped them producing sperm
have fathered offspring after a
team in China fixed their infertility
by editing their genome.
The approach, which used the
CRISPR DNA-editing technique,
could one day help many infertile
men around the world. “Most
of us think this is the future for
these men,” says Sarah Vij,
an infertility specialist at theCleveland Clinic Foundation in Ohio.
There are several known genetic
mutations that prevent the stem
cells in the testes giving rise to
healthy sperm. For decades,
biologists have been exploring
ways of correcting these mutations.
In 2015, for instance, one team
showed that they could restore
the ability of mouse stem cells to
produce healthy sperm by using
CRISPR genome editing to correct
the underlying genetic mutation.
Now Xiaoyu Li’s team at the
Affiliated Hospital of Guangdong
Medical University has pushed this
work a step further by taking stemcells from a mouse, correcting the
mutation and implanting them back
in the same mouse. Four months
later, the mice mated with females,
and nine out of 11 fathered healthy
offspring (Stem Cell Research
& Therapy, doi.org/c9zz).
However, it is too soon to try this
in people. The biggest obstacle is
that there is no reliable way of
isolating human sperm stem cells,
says reproductive expert GeertHamer of the University of
Amsterdam in the Netherlands.
Vij thinks using germline genome
editing – DNA changes that can be
passed on to offspring – to treat
infertility would be justified. But
the birth of many children as a result
of using CRISPR would also open
the door to other uses, she says.
“It’s a slippery slope, right?
All of a sudden you are doing
a tremendous amount of
gene-editing, creating scientifically
made babies,” Vij says. “Once you’re
able to do this, then you are able to
edit for all sorts of things.” ❚
Michael Le PageArtificial intelligenceDonna LuAI learns to deblur faces
Method may help identify people from low-quality CCTV images
Poor images may no longer
be a barrier to identifying
suspects caught on camera