Daguerreotypes start as copper plates with a thin, mirror-polished coating of silver that’s been exposed to halogen gas (iodine or bromine) to make silver halide. Light hitting this compound knocks an electron loose, which attaches to a silver ion, forming a neutral silver atom. The result is that all the places on the plate exposed to light are clusters of pure silver, and the rest is silver halide.

Next, the exposed plate is held over a warm pool of mercury (don’t breathe!). The mercury combines with the silver atoms, creating the equivalent of a digital image’s pixel: a tiny “grain” between 150 and 800 nanometers in diameter that scatters light, making areas of the surface that were exposed to more light appear brighter. Finally, the plate is soaked in sodium thiosulfate, which washes away the unexposed silver halide, leaving dark regions — the image’s blacks and grays.

The result is a one-of-a-kind direct positive — as opposed to the negative produced by modern chemical photography — with a haunting, soft, almost three-dimensional quality. Look at a daguerreotype from the wrong angle and you’ll see only a reflection: The image is trapped inside the mirrored surface.

The surface of a daguerreotype is so finely textured that dust can adhere to it and become impossible to remove without damaging the image. And even the tiniest particles, which at 30X to 100X magnification show up as bright white spots, can sometimes cause the surrounding areas to tarnish.

That’s where the computer vision specialists came in. A team led by Ross Messing, Paul Ardis, and Xiaoqing Tang found that in the Cincinnati daguerreotype, about one pixel in 500 was obscured by dust. And with the series totaling nearly 9 billion pixels, identifying and repairing all the problem areas by hand was obviously out of the question.

Instead, they marked all the dust specks in one small patch and then used machine-learning techniques to train a computer to detect others automatically. The computers searched for other regions in the image similar to the surroundings of the missing pixels and then performed simple copy-and-paste operations to fill in the holes.

As a historical record, the Fontayne-Porter daguerreotype is unparalleled. It contains the first photographic images of steamboats, a railroad station, and one of the country’s earliest astronomical observatories. It may also be one of the earliest pictures to show free blacks, who were building a community in Cincinnati, just across the line from Kentucky slave country. A ditch running from the corner of a building down to the river — eroded by effluent from an outhouse — presages the cholera epidemic that hit the city the following year.

Even artifacts of daguerreotype preparation yielded new knowledge. The silver surface of an unexposed daguerreotype is tricky to polish to a mirror finish — even the finest cloths or brushes leave tracks that are clearly visible at high magnification. But the art historians didn’t want those marks removed; they wanted to be able to enhance them. It turns out that the streaks act as signatures. Each daguerreotypist had a distinct method of polishing — sweeping tiny suspended brushes across the plate or hand-polishing (as Fontayne and Porter did) with carefully chosen cloths. The resulting patterns vary, but in a small region they all look like very fine, roughly parallel dark lines. So Messing, Ardis, Tang, and their collaborators designed an algorithm to detect these unique patterns and bleach out the rest of the image.

After all the restoration, historians now know the exact hour and minute when the image was captured. Back in 1947, steamboat enthusiast Frederick Way and Cincinnati Public Library director Carl Vitz undertook an extensive historical investigation of the daguerreotype, using steamboat records to identify the only date on which all of those vessels were in Cincinnati: September 24, 1848. And by analyzing the angles of shadows, they figured the shots must have been taken just before 2 pm. A clock tower showed the time, but however much the researchers strained to read the 1-millimeter-diameter clock face with a magnifying glass, they couldn’t make it out.

After the images emerged from Eastman House’s microscope scanner, the team cheered when they saw the clock tower: It read 1:55.