Connecticut Wrote Six Rules for Photoshopped Evidence in 2004 and Kept an Innocent Man in Prison Anyway

A judge in Hartford wanted to know how the computer had made a set of plaster teeth semi-transparent so they could be laid over an autopsy photograph of a murdered woman's breast. "Judge, you got me," the state's forensic odontologist answered. "I don't know how they do it, but they do it," he said, and then, "I think we really need an engineer."

The witness was Constantine Karazulas, chief forensic odontologist for Connecticut's state police laboratory, and he had told the jury that Alfred Swinton made those marks with reasonable medical certainty and without any reservation. The jury returned a murder conviction on March 21, 2001, and the sentence was 60 years. About a month later, and before Swinton had been sentenced, the New Britain company whose software had enhanced the bite mark photographs issued a press release announcing that its new image-enhancing product had helped solve a 10-year-old case and that the man convicted faced 60 years in prison.

When the appeal reached the Connecticut Supreme Court three years later, the court took up a question it had never faced before and that few American courts had addressed at all, which was what a party must prove before a digitally enhanced photograph goes to a jury. State v. Swinton, 268 Conn. 781, produced a six-part foundation test for computer-generated and computer-enhanced exhibits that courts and evidence commentators still reach for. It is a careful opinion. It is also the opinion that kept an innocent man in prison for another 13 years.

What Lucis Did to the Photograph

The enhancement software was called Lucis, built by Image Content Technology of New Britain, Connecticut, and developed in 1994 for scientific applications. The state introduced its output through Major Timothy Palmbach, who oversaw the division of scientific services in the state department of public safety and held a master's degree in forensic science. Because the state police lacked a high-resolution scanner and printer, Palmbach produced the trial exhibits at the software maker's own offices, with Karazulas standing over his shoulder telling him when the image looked right.

Palmbach's explanation to the jury is the clearest description of local contrast manipulation you will find in a criminal appeal. An ordinary photograph, he said, carries around 255 layers of contrast, and the human eye can perceive at best 32 of them, so most of what is recorded in an image is never seen. Lucis let him select a narrow band of those layers and suppress the rest, which pulled faint detail forward. Push it too far on skin and the pores, hairs, and wrinkles all arrive at once, so a second control dialed the ultrafine detail back down. The underlying algorithm, differential hysteresis processing, was patented in 1998 with Klaus-Ruediger Peters named as inventor and the University of Connecticut as assignee.

Palmbach ran the whole process on a laptop in front of the jury, side by side with the unenhanced original. Asked whether the enhanced image contained anything the original did not, he said no, and explained that Lucis does not remove pixels, only selects which range to display. Every bit of contrast was still there, he testified, some diminished and some brought forward, purely for viewing. He had also tested the software the way any careful photographer would test a new tool, by photographing a bite mark that Karazulas put on his own arm and comparing the enhancement against the original.

The court went further than the state had asked, and then accepted the foundation anyway. Lucis reveals detail to the human eye that was not visible before, the justices wrote, so the program actually is creating some part of the image, to the extent that it reveals what was not visible before. The state had argued the enhancements were just reproductions. The court said no, treated them as computer-generated evidence, and applied a stricter standard anyway.

The Six Factors the Court Borrowed

Having decided that a contrast adjustment counted as generating evidence, the court needed a test. It went to Rule 901(b)(9) of the Federal Rules of Evidence, which requires a party to describe a process or system and show that it produces an accurate result, and then lifted a six-item list out of a Mueller and Kirkpatrick evidence treatise. The proponent has to establish "(1) the computer equipment is accepted in the field as standard and competent and was in good working order, (2) qualified computer operators were employed, (3) proper procedures were followed in connection with the input and output of information, (4) a reliable software program was utilized, (5) the equipment was programmed and operated correctly, and (6) the exhibit is properly identified as the output in question."

A 2007 Stetson Law Review article gave the list a name that stuck, calling it the Swinton Six. The court was careful to say these were guideposts rather than a mechanical test, and it declined to bless or condemn any particular program, on the sensible ground that whatever software it named would be obsolete before the ink dried.

Lucis cleared the factors the court worked through. Palmbach testified that the program was relied on by pattern analysts for fingerprints, bloodstains, footwear and tire impressions, and bite marks. He was qualified in pattern evidence and had produced digitally enhanced images of dental impressions on skin twice before. He walked through digitization and layer selection on direct, on voir dire, and on cross. He noted that Lucis, unlike Photoshop, had no image editing features at all. The one gap, that he had never seen a published error rate for the software, did not sink it.

Where the Photoshop Overlays Fell Apart

The second exhibit set was different. Karazulas wanted images of Swinton's dentition superimposed on the bite mark photographs, some as acetate-style tracings and some as translucent scans of the dental casts themselves. He could not do it. He recruited Gary Weddle, a Fairfield University chemistry professor, and spent roughly seven to eight hours over two days watching Weddle build the overlays in Adobe Photoshop. Weddle never testified, and the opinion notes that the record does not reflect why.

So the man on the stand explaining the most visually persuasive exhibit in the case admitted he had "no skill or experience" with the software that made it. Asked how the teeth were made translucent, Karazulas said they would have to ask a Photoshop expert or a technician, though it made sense when you watched it happen on screen. Asked how the superimposition itself worked, he said Weddle moved them together. Asked how he knew the images had not been altered, his answer was that he had asked Weddle not to change anything.

Palmbach, called briefly on the point, testified that Photoshop "certainly has features where you can take multiple images and bring them in and out of plane in an attempt to overlay them over each other or rotate them and try to align them." That is exactly the capability the defense was worried about. Defense counsel called it a black box phenomenon and argued that two authenticated inputs do not guarantee an authenticated output, because a scan is a transformation and an alignment is a decision. The court agreed and held the Photoshop overlays improperly admitted.

Then it held that the mistake was not a constitutional one. The defense had its own forensic dentist, Neal Riesner, who used the state's overlays to argue the opposite conclusion, pointing to teeth with no corresponding marks and calling the layer selection a "trick of the computer." Riesner's ability to testify from those overlays, the court wrote, "undermines the significance of the defendant's inability to question Weddle," so the error was evidentiary rather than a denial of confrontation.

Then it held the evidentiary error harmless, and that turned on considerably more than the pictures. Karazulas had reached the same conclusion in 1998, before either program touched the case, at the same stated level of certainty, and the court found his identification rested on properly admitted exhibits, with the overlays serving "merely to augment the certainty of his conclusion." From there it weighed the rest of the state's case: the victim's black bra and the trash bags found in and around Swinton's apartment building, the safety pins from his van, and what the opinion calls "the numerous confessions and other incriminating statements the defendant made to several people."

The Question That Was Never on Appeal

The popular version of Swinton, the one that shows up in evidence outlines and continuing legal education slides, is a story about judicial rigor. A court looked hard at digital manipulation before most judges knew what a layer was, and wrote a durable rule. That much is true. The line it drew was not the one the summaries report, though. The court refused to treat Lucis as mere redisplay and held it to the computer-generated standard anyway. What separated the two exhibit sets was the witness. Palmbach could establish equipment, operator, procedure, reliability, and operation for Lucis. Karazulas could establish none of them for the overlays, and the man who actually built them never took the stand. That Photoshop was "capable of actually altering photographs" appears in the opinion prefaced by "Moreover, we note." It is an observation, not the ground of decision.

What the opinion never asks is whether the comparison those images enabled works at all. Two forensic dentists looked at the same overlays and reached flatly opposite conclusions, one describing fifteen unique characteristics and the other testifying that several of those exhibits showed no concordant points at all. That is not a foundation problem. It is a validity problem, and it was not raised. Joette Katz, who wrote the unanimous opinion, said so plainly years later: whether bite mark analysis is a sound forensic method capable of identifying a suspect was simply not an issue on appeal.

The science had already been shaky when Swinton was tried, and it collapsed afterward. A 2009 National Academy of Sciences report found that bite marks on skin change over time and are distorted by the elasticity of the skin, the unevenness of the bitten surface, and swelling and healing, and concluded that "the scientific basis is insufficient to conclude that bite mark comparisons can result in a conclusive match." The uniqueness of human dentition, the report noted, has never been scientifically established. Its own professional body, the American Board of Forensic Odontology, has since withdrawn permission for analysts to declare a positive identification from a bite mark at all.

What the DNA Found

At the request of the Connecticut Innocence Project, swabs taken from the bite mark were tested in 2014 and 2015. They carried male DNA that was not Swinton's. Testing of the black bra recovered from his building's basement turned up skin cells matching neither Swinton nor the victim, which suggested the victim had never worn it. Vaginal and anal swabs excluded him. So did scrapings from under the victim's fingernails.

Karazulas then signed a sworn affidavit, the kind of recantation a forensic expert almost never gives. It stated that "new scientific developments caused a significant change to my understanding of bite-mark analysis, which have compelled me, as a matter of my professional ethics and civic duty, to recant completely the testimony that I gave at Mr. Swinton's trial." Skin, he wrote, "is not capable of accurately recording the characteristics of human dentition, making it impossible to identify 'the biter' with any degree of certainty."

Hartford State's Attorney Gail Hardy agreed the conviction should be vacated. Swinton walked out on bond on June 8, 2017, having been in custody since his rearrest in October 1998. Prosecutors dropped it entirely the following March. According to the National Registry of Exonerations, he received $3.6 million from the state in 2020.

Enhancement, Alteration, and the Log You Keep

Every photographer carries an intuition about where the line sits. Recovering shadow detail that the sensor already captured is not the same as painting in a highlight. Local contrast, whether you get it from a curve, a clarity slider, or careful dodging and burning, redistributes what is in the file. Cloning a distraction out of a frame does not. Swinton is the rare document where a court tries to write that intuition down, and it gets remarkably close before it stops.

The case turns on the person, not the pixels. Palmbach's enhancements survived because he could sit there and account for every step, on a laptop, in front of the people who would decide. Karazulas's overlays failed because the man presenting them had watched someone else click a mouse for eight hours and could not say what any of the clicks did. Keep the original file untouched, keep a record of what you did to the copy, and be able to narrate it. That is not a courtroom rule. It is the same discipline that lets you answer a client who asks whether the sky in the frame is the sky that was there.

None of it saved Alfred Swinton, because a foundation test asks whether an image faithfully shows what the process produced, not whether the process was measuring anything real. Joette Katz, who wrote the six factors that outlived the conviction they upheld, later wrote in the Connecticut Law Tribune: "My regret at the loss of years of freedom Swinton experienced is palpable."

Lead image by Kenneth C. Zirkel, CC BY 4.0. Source.