Semi-Legibility and Visual Evidence: An Initial Exploration

A group of Hmong teenage boys were biking near an elementary school in North Minneapolis, in the summer of 2006. A patrol car spotted them and suspected they might be making mischief, though the police hadn’t seen them break any law or even do anything overtly suspicious. The police hovered nearby, then followed as one of the teenagers tried to depart, first on his bike, and then by foot. After a short pursuit, a policeman shot 19-year-old Fong Lee nine times at close range, hitting him with eight of the bullets. The young man died at the scene. The police insisted the shooting was justified. The officer said he saw the teenager carrying a gun and he refused to drop it notwithstanding police orders to do so. The officer reported that when he shot Lee, a weapon flew from his hand as he fell, landing a few feet from his fallen body. The young man’s friends, however, all insisted that Lee had not been carrying a gun. The Russian semi-automatic pistol found on the ground by his body lacked fingerprints or any other forensic evidence to link it to Lee. Furthermore, some evidence suggested that the gun police claimed to find a few feet from Lee’s body might already have been in police custody. The boy’s parents cried foul. Their son’s alleged weapon was a plant, they insisted, left by the police to justify an inexcusable shooting. In the excessive force trial against the police, the plaintiff introduced surveillance videos from the police car’s dash-cam and from school cameras that had caught much of the incident on tape. Unfortunately, the videotapes were of less than stellar quality, blurry and hard to interpret in key particulars. The policeman’s gun is easily seen on the video, and was clearly drawn and pointed straight at the boy as they both ran across the school’s grounds. But was Lee himself also carrying a gun? Was anything at all in the boy’s hands as he ran from the police? What did the video show – was that an edge of a shadow, or was it nothing, just a trick of the light? Were his hands empty, or could that possibly be an object, perhaps even a gun? Though the plaintiff tried to argue that the video provided “conclusive evidence” that Fong Lee carried no gun that fateful day, ¹ the video was just too fuzzy to answer that key question with certainty.

A 19-year old driver was clocked going 73 miles an hour on a four-lane road, late at night, and failed to stop for the police. The cops pursued the driver through a shopping center and back onto two-lane roads, at speeds up to 85 mph. After six minutes and ten miles of driving, the young driver was chased off the road and down an embankment, as one of the police cars pursuing him purposefully rammed him from behind. The driver survived, but was left a quadiplegic. He sued the officers, alleging a violation of his civil rights through the use of excessive force by the police. The lower courts refused to grant the police summary judgment, but a majority of the Supreme Court deemed summary judgment quite appropriate. ² They thought that no reasonable dispute of fact existed, primarily because of the footage from the video cameras lodged in the dashboard of the police vehicles, which provided what they took to be compelling visual records of what had transpired that dark evening on the roads of Croweta County, Georgia. For Justice Scalia, who penned the majority opinion, the video evidence literally proved the defendant’s case; he declared that he was “happy to allow the videotape to speak for itself,” and hence it was posted on the Court’s website so readers could view it themselves, marking the first time in history a video was provided along with an opinion from the Supreme Court. ³ What, according to Scalia, did the tape show? He sees “respondent’s vehicle racing down narrow two-lane roads in the dead of night at speeds that are shockingly fast,” as it “swerve[s] around more than a dozen other cars.” It was “a Hollywood-style car chase of the most frightening sort, placing police officers and innocent bystanders alike at great risk of serious injury,” and “so utterly discredit[ing]” the plaintiff’s version of events that his story warranted no credence, notwithstanding the deferential standard applied to the non-moving party’s version of facts in a summary judgment motion. ⁴ But Justice Stevens, in a lone dissenting opinion, saw matters – and the video record – quite differently. He saw no evidence that the plaintiff ever lost control of his vehicle. He saw relatively empty roads, no pedestrians, and based on the video, was not confident that the defendant’s driving posed a serious danger to others that warranted the actions taken by the police. He thought his fellow Justices were sometimes misinterpreting what they saw on tape, often overreading the video’s implications, and most fundamentally, failing to understand that it was capable of multiple plausible interpretations, and that summary judgment was therefore inappropriate. ⁵

Levi Bellfield was on trial for the grisly murder of two teenage girls and the attempted murder of a third, and police suspected he was a serial killer linked to numerous other as-yet-uncharged killings as well. One of the girls was bludgeoned to death on her short walk home from the bus stop. CCTV captured numerous images of a car overtaking her bus, believed to belong to the perpetrator. Despite the presence of eight cameras on the bus, the only captured images of this vehicle were low-resolution, nighttime images taken in “time lapse” mode; essentially, a series of blurry still photographs rather than a continuous moving image. One single frame captured the license plate of the car, but though dark colored fragments were visible on the plate, not a single digit or letter on the plate was visually clear enough to be easily deciphered. The prosecution alleged that the car in the images was a Vauxhall Corsa, and that the blurry image of the license plate in fact read Y57RJU, the license plate of the defendant’s automobile. The CCTV images were introduced in evidence, along with the testimony of forensic imaging experts, who concluded that the blurry images indeed showed a Vauxhall Corsa. While these experts acknowledged that the license plate image was too distorted to make an absolutely certain identification of each letter and number, they examined each character on the image, assessed what characters were, in their view, plausible given what they could observe, and concluded that the visual information was consistent with Bellfield’s license plate. ⁶

A baby, less than one year old, was rushed to the emergency room, barely able to breathe, with a suspected brain injury. The infant survived, but suffered permanent injuries so severe that he would be unlikely ever to walk, talk normally, or feed himself. The diagnosis given was “shaken baby syndrome,” and the mother’s boyfriend was soon charged with felony child abuse. Some of the key evidence in the criminal case included the CT scans and MRIs taken at the hospital, which showed the presence of severe brain injuries. The physicians and radiologists called by the prosecution testified that these images revealed the presence of subdural hematomas and brain swelling. They also testified that these injuries were very probably due to someone’s having shaken the baby, hard. The radiologists’ reports, and likely the CT scan and MRI images themselves, were introduced as exhibits for the jury, along with the testimony of three physicians. The defendant unsuccessfully challenged the admission of each expert’s testimony. Among other arguments, the defendant suggested that the radiologist was qualified to interpret the medical findings in these high-tech tests, but not to speculate about the underlying causes of those medical findings – in other words, to describe the medical findings shown by the scans, but not to attribute those findings to a specific mechanism like shaking. Both the trial court and the appellate court disagreed. The jury agreed with the physicians’ interpretation of the findings on the medical scans, and the defendant was convicted. ⁷

An 84-year-old woman was beaten and stabbed to death in her home back in 1975. No suspect was identified and the case was closed unsolved. Eighteen years later, a fingerprint from the scene was run through a database and produced a “hit,” an apparent match to someone’s print. Confronted with the fingerprint evidence, that suspect confessed to involvement and implicated two other co-conspirators: her long-ago boyfriend and the defendant. After that suspect’s confession, her boyfriend was located in state prison and interviewed (but not charged), but he provided no information about the defendant, who was neither located nor charged at that time. Another 15 years passed, and in 2008, another, different fingerprint lifted from the crime scene was matched to one in a national database, this time to the defendant. By this point, the third alleged co-conspirator, the woman’s boyfriend, was dead. The defendant stood for trial, now more than 30 years after the alleged crime. The evidence against him consisted of the pair of crime scene fingerprints (one identified as his accomplice’s and the other as his), along with the testimony of his supposed accomplice, whose credibility was rather dubious, given that she had a substantial history of mental illness, including hallucinations, and was currently on probation for a felony perjury charge. The fingerprints were therefore clearly the crux of the prosecution’s case. Two experts testified in detail about the fingerprint matches. Though the appellate opinion is not explicit on this point, the experts most likely, as is typical for fingerprint experts in significant criminal trials, presented enlarged and annotated charts to the jury. These visual displays would have depicted the various points of similarity among the friction ridge minutiae that undergirded their conclusion that the crime scene print at issue could only have come from one and only one possible source: the defendant. ⁸

We have here five vivid examples of visual evidence playing a significant role in the trial process, but beyond that, what do they have in common? The first is filmic evidence, depicting in full-color and motion, if not nearly as crisply as a viewer might wish, the key events underlying a civil suit. The second also involves filmic evidence recording critical moments at issue in the case, and which some interpreters, including a majority of the Supreme Court, thought spoke for itself, while others – Stevens, in dissent – deemed the film’s meaning far less clear. The third is a set of blurry photographic still images snapped by a surveillance camera, a possible circumstantial link between the perpetrator and the victim in a sensational criminal trial. The fourth is neither a film nor a photo, but the visual displays produced by a high-tech scanning device primarily designed to assist with medical diagnosis by providing visual depictions of the human body in ways otherwise unavailable. The final example captures crime scene evidence invisible to the naked eye, rendered visible through the labor of a crime scene investigator, brought into court, decades later, blown up and annotated to illustrate its match with the defendant’s fingertips. The examples all involve visual evidence of some kind or another, but beyond that, what unites them?

I want to suggest that all five of these examples of visual evidence share a key quality: every one of them is, at least in part, semi-legible to a lay observer. If an ordinary observer were to examine the films, the still images, the MRIs, or the fingerprints, she would likely believe she could glean some useful information from it. But the observer of any of these images (with the possible exception of the films in Scott v. Harris), would also, simultaneously, very likely recognize her limited capacity to interpret and unpack that information, because the image is blurry, ambiguous, or requires interpretive expertise that she doesn’t have. She would know that she can only partly make sense of what she sees.

Semi-legible visual evidence is, as its name suggests, neither entirely legible nor impossible to read. It is neither fully opaque nor fully transparent, neither jibberish nor self-evident. Semi-legible visual evidence does not indisputably mean something in particular, but it is not so capacious that it can mean anything at all. Gleaning meaning from it, making it more legible than it first appears, requires the exertion of effort, expertise, or both, and even with such effort and expertise, the visual evidence at issue may still remain subject to question, doubt, and ambiguity, capable of being understood differently by others.

This notion of semi-legibility, I argue, is a useful lens for analyzing visual evidence. Within its ambit I include not only films and photographs but other kinds of visual depictions and inscriptions made use of in legal settings, including, for example, the many kinds of body scanning technologies whose images are often used in cases ranging from medical malpractice to criminal law; and the numerous forms of pattern identification evidence that are a mainstay of forensic science. This notion of semi-legibility usefully focuses our attention on the ways that much visual evidence neither speaks for itself nor permits unbounded interpretations, but rather, has a range of plausible – and potentially inconsistent – readings. It is this push-pull quality, this doubleness – the forceful constraints these forms of evidence do impose upon the plausibility of interpretation, simultaneously coupled with the inevitable remaining legitimate multivalence of meaning – that makes these forms of visual evidence complex to understand, both procedurally and substantively, as forms of legal evidence. Visual evidence, from photographs to films to CT scans, fingerprints and x-rays, can be, on the one hand, both indexical and evidentiary – a “mirror with memory” ⁹ – and yet nonetheless, contra Justice Scalia, not simply speak for itself. Thinking about such evidence as semi-legible helps, I hope, to keep both of these issues in the picture. Recognizing “semi-legibility,” in other words, invites us to look at the strategies, techniques and even the legerdemain by which those who make use of visual evidence – from the lawyers who introduce it, to the experts who explain it, to the jurors who assess it – work to read meaning into ambiguous evidence and make it into proof.

The word legible, of course, is etymologically derived from the Latin verb legere, “to read.” While it is linked, in typical usage, to ideas of the visual, to what can be clearly seen – and hence read – it is most regularly used in reference to text, to the act of being able to read and distinguish letters, rather than applying to graphically more complex visual stimuli. We more customarily talk about whether an item of writing is legible than whether, say, a painting is. But just as a person’s poor handwriting can make their script difficult to decipher, a semi-legible piece of visual evidence cannot be completely deciphered, fully understood, unequivocally “read” or interpreted. Legibility, as I am using the term, is therefore partly metaphoric when applied to non-textual visual evidence, the borrowing of a concept from textual assessment and applying it to visual depictions more generally. But it is only partly metaphoric. For when visual displays are used as evidence, as a form of legal proof, they are, to a great extent, being introduced in evidence precisely in order to be “read.” The profferer wishes the factfinders to find some kind of factual information, knowledge, or meaning about what was, in what is displayed; they are supposed to be able to read something about the state of the world via the image. Their ostensible purpose is primarily interpretive rather than aesthetic, or to put it crudely, to assess truth rather than beauty. ¹⁰

To make out semi-legible writing, a reader needs to look carefully, focus, examine and re-examine, perhaps even squint. A reader may use clues derived from the more legible letters to make reasonable inferences about letters that are less so. Seeing the word “doc?or,” the reader may decide that the illegible letter is very likely a “t”; if the assessment of the rest of the letters is accurate, there are, as any dictionary will indicate, no other English language words that fit. Or a reader may use contextual clues to interpret the text in light of what would make sense. In the sentence, “I’d been ill for several weeks, so I made an appointment with my d????r,” the reader may infer that the illegible word is doctor, not because there aren’t other possible interpretations for that word standing alone (drawer? dealer? dancer? to name just a few possibilities) but because, in context, one option “fits,” interpretively, better than the others with the expected meaning of the sentence or paragraph. Or a reader may turn to someone with specialized knowledge for help in deciphering the illegible letters: perhaps a family member or close friend of the author, with experience interpreting that particular writer’s scribbles, will be better at understanding what was written than a stranger or a mere acquaintance. In some instances, professional experts on handwriting assessment might assist the interpretation; or perhaps technology will help, from basic photographic enlargement to more sophisticated digital enhancement techniques.

We thus see at least five strategies or techniques for trying to decipher illegible words: (1) lexicographic gap filling, where the legible letters are used to glean information about the intelligible possibilities for the illegible ones; (2) semantic, or contextual gap-filling, where the meaning is inferred from context not on the basis of the surrounding letters in the same word, but at a somewhat broader level, by using the rest of the sentence, or paragraph, to assess what best makes sense; (3) using the specialized knowledge of those with relevant personal experiences, like the author’s spouse or mother, to evaluate the uncertain letters; (4) using the specialized knowledge of someone with expertise not in this person’s writing, but in handwriting assessment in general; (5) using technological enhancement techniques to increase the visual clarity of the specimen of writing.

Depending on the circumstances, any or all of these interpretive strategies may be helpful for deciphering intelligible text. At the same time, each and every one of them may also generate risks for possible misinterpretation or for cognitive bias to influence judgment too. Perhaps the word that seems the best “fit” isn’t actually correct – the writer could have been saying something odder and less expected than “doctor,” like, “I had been ill for several weeks, so I made an appointment with my dagger.” Or perhaps erroneously interpreting one word leads the reader to mistake a subsequent one (if the reader is wrong about the word “ill,” she’s more likely to be wrong about “doctor” too). The writer’s mother might, notwithstanding her familiarity with the author, erroneously interpret the letters at issue, and so might the handwriting expert. And technological enhancement might not make the letters any clearer, or might do so erroneously. This array of strategies offers possibilities for increased legibility but certainly no guarantees.

All of this largely applies to visual evidence as well. The point is not that these interpretive and inferential methods are failsafe, but rather that analogues of these strategies are possibilities (and risks) for the assessment of semi-legible visual evidence as well. Here too, there are a variety of deciphering strategies available, ranging from a focus on context and plausibility, to the use of experience and expertise, to technological enhancements, for increasing clarity. To be clear: I do not mean to suggest that there are not important interpretive and cognitive differences in the way we interpret text versus images; surely there are many. My point is merely that notwithstanding these differences, legibility, which connotes both reading and visuality at once, offers a useful conceptual lens for thinking about visual evidence and its use in court.

More broadly, my central purpose in this essay is to name and explore this category and quality of visual evidence – in other words, to stake out semi-legible visual evidence as a category worth identifying and exploring, and to recognize that a great many items of visual evidence made use of in the courtroom, from police cars’ dashboard videos to surveillance camera films to many of the forms of visual evidence regularly used by expert witnesses to explain, buttress, and ground their testimony, are indeed semi-legible, albeit to varying degrees. Second, I aim to illustrate that it is useful to think conceptually about multiple and disparate forms of visual evidence, forms often not engaged with together, in the same breath. While there has been much valuable work examining various specific forms of visual evidence – including filmic evidence, photographic evidence, fingerprints, polygraphs, PET scans, etc., ¹¹ – all of these forms of visual display share certain structural similarities, like semi-legibility, that are worth unpacking. Thinking about these different kinds of visual evidence together has the effect, I believe, not only of highlighting semi-legibility as a meaningful category, but also of suggesting often-unnoticed connections between expertise and visual legibility. ¹² By emphasizing the ambiguity present in these visual depictions, also invites a focus on questions of cognitive bias and its effects on perception and interpretation. ¹³

In what follows, I will describe semi-legibility in somewhat more detail, and then offer a taxonomy of various forms of semi-legibility that affect the use of visual evidence in court. I then turn to some reflections on this taxonomy, especially focusing on methods by which advocates may make semi-legible evidence seem more (or less) decipherable, and finally, offer some concluding thoughts on the complex connections between expertise, professional authority and visual knowledge in the courtroom.

How ought we to understand and analyze this idea of semi-legibility? First, we should recognize legibility as a matter of degree, rather than an all-or-nothing proposition. At one extreme we might have the depiction that seems to capture in extraordinary detail some perfectly fixed instant in time; at the other, we could have a visual display that literally approaches meaninglessness, at least to the lay interpreter. This quality is not an on-off switch, either legible or illegible, but exists on a continuum, and different parts of any given image, film, or other visual display may have varying degrees of legibility. In the surveillance camera video in the Lee case, the police gun is quite legible; the contents, if any, in the young man’s hands, are not. ¹⁴

Second, the legibility of visual evidence is also relational, depending not just on the image itself, but on what questions are being asked of the image and what else the viewer knows. An image may be highly legible vis-à-vis some matter of relevance, and, simultaneously, visually difficult to parse in relation to a different question. In the time-lapse photos in the Bellfield case, if the question at issue was: “was there or was there not a car located behind the bus at a certain point in time?” it is no challenge to “read” the image and provide an answer with which any reasonable viewer would agree. If the question is whether the car at issue is a Chevy or a Vauxhall, it is far harder to read the image for this information.

The relational dimension of legibility operates not only with respect to what is being asked of the image, but also in relation to who is viewing it. Images may be differently legible to different viewers, depending on their background knowledge, experience, and expertise. Just as my poor handwriting may be deciphered easily by a close friend with substantial familiarity with how I form my letters, and yet remain inscrutable chicken scratches to an acquaintance unfamiliar with my messy hand, the legibility of an item of visual evidence may vary significantly depending on the viewer. Some issue of relevance might be easily legible to someone with geographical knowledge of the streetcorner being depicted on a video, or to an expert with a certain kind of interpretive know-how, yet remain opaque to someone lacking that knowledge. To give a concrete example, consider the videos shown in Scott v. Harris. ¹⁵ The majority opinion thought that the video simply spoke for itself, and that any reasonable interpreter would view it in a similar way. The dissenting opinion, by contrast, emphasized that local knowledge of the streets of Georgia, or experience having learned to drive on two-lane roads rather than superhighways, might legitimately affect how a viewer understood and interpreted what the dash-cam videos depicted. The basic insight of hermeneutic theory – that understanding is dialogically produced by both the matter being interpreted and the interpreter – surely applies to visual evidence as much as text. ¹⁶

Third, legibility is related to interpretability, but it is not precisely the same thing. I do not mean to devolve all issues of image interpretation into issues of visual legibility. A semi-legible image clearly requires interpretation, but the fact that an image requires interpretation does not necessarily mean it is semi-legible. Even the most legible and seemingly transparent of images requires interpretation; furthermore, highly legible images may nonetheless mislead a naïve observer. Indeed, visual clarity itself may be misleading: what looks clear in the image may have been far less transparent to an actual eyewitness observer, due to the perspective, shadows, or color differences in the image compared to the actual world, or as a consequence of the techniques of image production. Or the visual point of view and frame of reference of the depiction may draw more (or less) attention to an object than it might have had if seen live and in person. Furthermore, a given image might actually have been staged, posed or altered rather than simply captured, and the viewer may not have been told or adequately understood this point. More generally, even when images are not posed or altered, the circumstances and methods surrounding an image’s production and editing may not be known or appropriately assessed by the viewer. Both of these kinds of issues – the important dimensions of the camera’s point of view in framing and shaping what is depicted, and the specific issues surrounding the circumstances of a particular visual display’s construction – may matter a great deal to its appropriate interpretation, but I think they operate in a register rather separate from my focus here on legibility. ¹⁷

All visual evidence, however legible, does require interpretation. I take as given that visual evidence, even those unstaged, unposed, unaltered photographic and filmic depictions of the world around us – what Jessica Silbey has felicitously called “evidence verite” ¹⁸ – do not simply provide an objective window on the world. Even these forms of evidence have a point of view; a place from which they were taken; a perspective; and are mediated and shaped by the particular forces – both human and technological – used to create them. Moreover, even these powerful and compelling visual depictions can be embedded within and framed by other narratives, stories, and evidence, and these captions, stories and perspectives can meaningfully shape how the seemingly objective visual evidence is read and understood. (Think Rodney King.) ¹⁹

At the same time, while visual evidence does not tell a single, unimpeachable story, the range of available interpretations for any given visual depiction is bounded, limited and constrained. No one could look at the Lee video and reasonably claim that Lee stopped promptly for the police when they first requested him to, or that the pursuing officer was unarmed. The range of plausible understandings is framed and limited by what is clearly depicted, and the legibility of any visual evidence – understood in relation to the questions being asked of it and the background knowledge and experience of the viewer – is part of what constrains and limits the range of possible interpretations.

Is all visual evidence semi-legible? Yes, probably so, at least to some degree – but for the purposes of this essay that question can be largely sidestepped. Legibility exists on a continuum, and my focus here is primarily on images on which there would be a likely consensus that legibility is limited. Whether we label those images at the outer boundaries of the category as completely legible, or just more so than most, is less important than recognizing the existence of the range. (Here again, Scott v. Harris is a kind of limiting case, as in that instance, the question of legibility was precisely what was in dispute.)

While it is valuable to recognize the category as a whole, we may also understand it better by breaking it down into subcategories. What then, are some of the specific ways that visual images may be semi-legible? How might we categorize the possibilities? I offer here a basic taxonomy of five categories, while recognizing that many instances of semi-legible evidence may fall simultaneously into more than one. My basic taxonomy proceeds as follows: (1) blurry, indistinct, or visually ambiguous images; (2) interpretively ambiguous images; (3) “jigsaw images” – where one or more important pieces of information are missing; (4) images semi-legible to laypeople or to those lacking relevant knowledge, experience, or expertise; (5) images semi-legible even to experts.

Blurry, visually ambiguous, or indistinct images: An image or film may simply be too blurry to be clearly seen. This could, of course, have many causes: perhaps it was taken by a low quality recording device; or under poor lighting conditions; or the camera shook or moved as it recorded; or the distance between the recording device and the subject matter of interest makes the object indistinct, and a close-up or enlargement becomes pixilated. Whatever the cause, some images are quite clearly visually unclear. The concept of semi-legibility is perhaps least metaphoric for this category of images. Indeed, in some instances, the need to parse blurry images can be literally a question of reading – consider, for example, the license plate image at issue in the Bellfield case, or a blurry image of a street sign in which the letters cannot be clearly made out. Both the Bellfield video and the Lee video are examples of this kind of visual ambiguity, where simply deciphering precisely what the images show poses a challenge identifiably separate from interpreting what those images mean more generally.

Interpretively ambiguous images: While they are, to be sure, closely related, there is a distinction between blurry images and interpretively ambiguous ones. To be more precise, the distinction is between ambiguity derived from a lack of visual clarity, and interpretive ambiguity that stems not from ambiguity at the indexical level of what is represented in the image, but from the ambiguity regarding the meaning of what is represented. Many images may be quite visually sharp and distinct, but subject to multiple interpretations, like Mona Lisa’s smile. In Scott v. Harris, for example, Justices Scalia and Stevens both agreed that the dash-cam video showed a number of cars pulled to the side of the road (no visual ambiguity there), but Scalia thought the video revealed that they were dangerously “forced” to the shoulder by the plaintiff’s driving, while Stevens thought they could equally as well have safely moved there in response to the police sirens’ warnings rather than the defendant’s driving. ²⁰ More generally, Scalia saw a “Hollywood-style car chase of the most frightening sort,” while Stevens saw the question of danger to other vehicles as far less obvious. ²¹ There may, perhaps, be more disagreement among viewers about the basic question of whether an image is ambiguous than about whether it is blurry. In an image taken in low light, or one that is pixilated or indistinct, viewers may disagree about precisely how much can meaningfully be seen and interpreted, but will likely agree on the basic fact of blurriness and the presence of visual ambiguity. By contrast, interpreters may disagree about whether interpretive ambiguity even exists in the first place: Justice Stevens saw the videotape in Scott v. Harris as ambiguous, while Scalia and the majority did not. Scalia thought reasonable people could only interpret the video in the same way that he did; Stevens vehemently rejected the idea that the video “spoke for itself.” ²²

“Jigsaw” images: This category consists of visual depictions with missing pieces, gaps, ellipses, or explicitly lacking relevant content. Images may be of extremely good visual quality, and yet fail adequately to show some matter relevant to the legal issue. The surveillance camera may pan away at a critical moment, or capture the image of the victim but not the perpetrator. Jigsaw images may be quite legible along some dimensions, but require inference, speculation and guesswork – or offer no meaningful clue whatsoever – regarding some other question of interest to the case. Jigsaw images remind us that with visual evidence, as with all evidence, questions of relevance and materiality are always relational: whether the evidence is more or less legible depends on what questions we ask of it, rather than operating as a quality that simply “inheres” to the visual display itself. In the Scott v. Harris case, Justice Stevens points out that while we know that two traffic lights were red when the police crossed the intersections, because we see the events from the police car’s perspective, we cannot know for certain that they were already red when the defendant crossed. That fact, certainly relevant to interpreting the degree of danger posed by the plaintiff’s driving, is not captured directly by the video, but requires us to infer something about missing information from what we can see. ²³ Similarly, in the Lee video, the dash-cam video from the police cars does not clearly depict the moment when Lee leaves his bicycle and sets off on foot, a moment of time important because the plaintiff contended that the police intentionally knocked against the plaintiff’s bike, scaring him (and therefore making him afraid to stop for the police), while the defendant denied ever being so close to the bicycle. ²⁴

Images legible to experts or those with specialized knowledge but semi-legible to laypeople: Many images may be only semi-interpretable to a layperson, but quite legible to an expert or someone with relevant specialized knowledge. A layperson without the relevant background and training cannot look at an FMRI, an x-ray, a fingerprint, or a polygraph, and simply know how to read it or how to interpret its significance. Some visual displays used by experts may be wholly opaque to lay viewers, as unreadable as a text in ancient Sumerian to an untrained modern viewer – but many kinds of visual displays created by and used by experts in court are semi-legible to many lay viewers rather than wholly incomprehensible. The degree of legibility will certainly vary from person to person, depending on the viewer’s prior exposure to such images and their interpretation, through whatever means, from personal experience, or from exposure through the news, pop culture, or from other media portrayals. Furthermore, the layperson’s beliefs about how to “read” these images may or may not be accurate. The point, however, is that for many forms of visual displays produced by experts, laypeople may still be able to glean something (semi-legible) from them as well. A layperson may well identify an x-ray image as an x-ray (as opposed to believing it an image generated by some other kind of visualizing technology), and if there is a human leg depicted on the x-ray, a layperson may be able to recognize that she is looking at the bones of a leg (as opposed to, say, a shoulder, a foot, or a dinosaur). Indeed, if the x-ray shows a severely fractured leg, a layperson may well “see” the break and, broadly speaking, understand what she is seeing even without expert assistance. But identifying an abnormal mass in a chest x-ray, or a greenstick elbow fracture with minimal dislocation, or interpreting a CT or PET scan or an FMRI, would likely be well beyond the skill of an untrained layperson. Depending on the circumstances, the same image might be straightforwardly “legible,” perhaps even obvious, for a qualified expert interpreting the relevant x-rays or body scan.

This category of semi-legible images does not necessarily require formal expertise or professional training for enhanced legibility. There may be images that are read quite differently by those with relevant specialized knowledge based on local knowledge or lived experience, rather than because of professional training. Someone who personally knows the location where a film was made may be able to identify particulars that a viewer without that real-life knowledge never could, to “decode” blurry or perceptually unclear visual images in light of their prior knowledge, or to better interpret ambiguous images. In this vein, Stevens, in Scott v. Harris suggested that those “more familiar with the hazards of driving on Georgia roads” would be far better interpreters of the video than the Court. ²⁵ Like professional expertise, relevant local knowledge or experience may improve a viewer’s ability to interpret correctly an interpretively ambiguous tape; these too are forms of expertise, though the relevant specialized knowledge derives from lived experience rather than formal training or the application of a self-conscious interpretive method.

Images only semi-legible even to experts themselves: In a certain sense, this is an extension of the previous category; the distinction is that some visual displays may be semi-legible to a layperson but (relatively) unambiguously interpreted by someone with the relevant expertise; whereas other images created by or used by experts may be semi-legible even to qualified experts themselves. Of course, one could drill down further and subdivide this fifth category along precisely the lines of the first three categories described above. To wit: the image’s semi-legibility could result from blurriness or lack of visual clarity, stemming, for example, from either the limited resolution of the imaging device, or the limited information contained in the object under study. Examples of expert images in this category could include a poor-quality latent print recovered from a crime scene, or the license-plate in the Bellfield case, in which experts claimed special, but still acknowledgedly imperfect, ability to interpret the blurry markings. Or, the semi-legibility could result from evidence that is visually clear but ambiguous in interpretation, like the question of the relationship between an abnormal brain scan and the behavioral meaning of that abnormal scan. The existence of an abnormality on the scan might be perfectly clear, as a visual matter, to a qualified expert, but whether or not that abnormality can appropriately be causally linked, for example, either to a specific psychiatric disorder or, even more specifically, to a criminal defendant’s specific behavior at issue in the case, might be highly disputable. The same applies equally to a victim: part of what was at issue in Middleton was whether the findings on the CT scan and MRI – which the defendant did not dispute – was necessarily caused by shaking the infant, or might have plausible alternative explanations. The question was not whether the images revealed subdural hematomas, but rather whether the subdural hematomas might have causes other than shaking. Or, the image may be semi-legible because of its “jigsaw” quality – part of what the expert would have liked to see may not be present in the image. A latent print, for example, may have high visual clarity but show such a small portion of the finger’s surface area, or a portion not present in the comparison print, as to make an identification impossible. Or a single x-ray, capturing only that moment in time, may make diagnosis ambiguous without an available comparison to a past x-ray (or other medical tests).

Focusing, as this taxonomy begins to, on precisely what makes any specific depiction semi-legible may help us better understand both how better to “cross-examine” any such evidence to make its possible limitations more salient, and, conversely, what strategies may make the evidence more powerful and persuasive. ²⁶ Note of course, that in an adversarial setting, both of these will typically be desirable (though to different parties): typically, the proffering party will wish to make the semi-legible image seem as persuasive as possible, while the opposing party will wish to make the limitations of the image as explicit as can be. Put simply, techniques that bring semi-legibility to the surface are likely to make a viewer more aware that the visual evidence does not simply speak for itself. Conversely, techniques that increase, or appear to increase, legibility, will often (though, as I will discuss below, not always) enhance the visual evidence’s persuasive power.

How, and to what extent, legibility can be increased, varies, of course, by category. For blurry images, possible forms of technological enhancement of the visual image may be of special focus; depending on the circumstances, interpretation that makes use of surrounding context information to interpret the blurry portion may also help. For interpretively ambiguous or jigsaw images, clarity may be enhanced by bringing the ambiguities or missing elements more explicitly to the surface, or by disaggregating the visual evidence, breaking the whole into parts. Delimiting the evidence may, potentially, permit the profferer to “isolate” the ambiguities, making the rest of the image seem more persuasive, or may permit her to suggest reasonable inferences for the missing jigsaw pieces based on what the neighboring puzzle pieces show. For evidence for which laypeople lack some relevant interpretive skill, enlisting expertise to interpret the image is an obvious strategy. By thinking of semi-legible evidence as a category, we can see that all four of these approaches – technological tinkering; using contextual information to permit inferences about what is missing; making ambiguities explicit; and using expert interpreters to guide the factfinder in “reading” the image – are all, though superficially quite different from one another, equally methods for increasing the legibility of the evidence presented. We see, too, the structural similarities between these strategies and those suggested earlier in this essay in relation to textual deciphering: there, too, we saw that turning to context (either the other letters of the specific word, or more broadly), or to experts, or to technology, all posed potential methods for making the written word more legible.

Moreover, questions about expert and technological authority and its relation to lay interpretation may arise with each of these methods, albeit in different ways. In this final portion of this essay, I look more closely at some of the ways these questions of expertise, technology and interpretation intersect for each of three categories: blurry images; ambiguous or jigsaw images; and images more legible to experts than to lay viewers.

Blurry images: For semi-legible blurry images, precisely what is the appropriate role for enhancement technologies and expert interpretation? To the extent that the enhancement techniques are framed as simply uncovering what is actually already there, just clarifying otherwise blurry “truth,” they may seem unproblematic. But to what extent has the techniques’ effectiveness at “clarifying” been tested, established and validated? Just how manipulable are the various enhancement techniques? Might they mislead the factfinder by creating a false and unwarranted sense of clarity? Precisely how much discretion and “authorial” power does a visual processing expert have, and how transparent is this power? Where is the line between enhancing an image and constructing a new one?

More broadly, whether through enhancement techniques or otherwise, to what extent can or should the factfinder legitimately be given expert assistance in interpreting blurry images? How ought we to think about the legitimacy of enhancement techniques vis-à-vis other possible sources of knowledge, and, especially, about the role of experts in enhancement or in image interpretation? ²⁷

Imagine, for example, a case in which a surveillance camera captures a convenience store robbery, but where the video is too blurry to make out details of the facial features of the gunman. Which of the following ought to be permitted at trial, and upon what preliminary foundation?

The testimony of a longtime acquaintance of the defendant, who identifies the blurry figure in the surveillance camera image as the defendant, because the figure depicted moves with the defendant’s gait and makes arm gestures just like the defendant. This is an expert with “local knowledge,” a form of case-specific expertise, though not an expertise derived from professional authority. Courts have long permitted lay identification evidence for voices, for handwriting, certainly for faces; is gait identification any different conceptually, and does it matter that the viewer is seeing the defendant on videotape rather than live? ²⁸

An expert who describes various image processing techniques and enhancements he conducted in order to make the images more clear, along with his conclusion, based on inspecting the enhanced images, that the defendant is (or is not) the person depicted in the image. Should the jury view the enhanced images for itself? Should the expert’s own opinion and conclusion also be permitted? Should enhanced images be permitted only if the originals are also introduced? In the Lee case, for example, the court permitted both the original and the enhanced images from the school’s cameras into evidence, but excluded the visual processing expert’s opinion about whether Lee carried a gun. Since the expert claimed his opinion was based on visual inspection the court reasoned that the jury could do that for itself. ²⁹

What if the expert claims that her visual inspection of the original (or, for that matter, the enhanced) images can be more sophisticated than the jurors? Should the court permit the opinion of an expert forensic image analyst with years of experience interpreting blurry images about what she has gleaned about the facial features of the person depicted in the blurry video? In the Levi Bellfield case, experts were permitted to testify about what they thought the blurry license plate showed. ³⁰ Should such experts need to establish that they truly do have a form of professional vision, that they can interpret blurry images more accurately and effectively than laypeople?

If enhancement techniques are used, how much leeway should the opposing party have to “cross-examine” the images and the techniques? Should they be able to illustrate the range of visual results that can be achieved through the use of enhancement technologies? Should they be able to introduce a “counter-expert”? Should they be able to introduce evidence about the dangers of visual suggestion and cognitive bias, and the risk that even an expert may suffer from confirmation bias, or the possible effects of having prior expectations about what the evidence would show? (In the Bellfield case, such evidence was indeed introduced, ³¹ though published opinions discussing the admissibility of such evidence are nonexistent.) Should the opposing party be able to “test” in court the visual acuity of the interpreting expert?

My purpose here is not to resolve these many questions, but merely to suggest their range. One point worth noting is that when we endeavor to clarify blurry images with the use of enhancement techniques, we create a significant link between this category of semi-legible images and those that are semi-legible to the lay viewer but (perhaps) not to an expert, in that questions of expert knowledge and its relation to visual transparency arise with particular force. How can and should expertise assist the lay viewer of a blurry image – is it by making the image itself more legible through technological manipulation; by the exercise of professional vision upon an otherwise semi-legible image; neither; or both?

Ambiguous or Jigsaw images: For interpretively ambiguous images and jigsaw images, parties may wish to engage in techniques that either showcase the ambiguities or the absent evidence or minimize them. Breaking down an apparently “whole” visual narrative into constituent parts may have a defamiliarizing effect on the juror, and may also make the viewing of violent or unusual events less emotionally-charged. That was certainly the defense’s approach in the Rodney King case, in which the apparently brutal beating was examined bit-by-bit in the service of an argument that what at first looked like obvious police brutality became far less legibly so when analyzed moment-to-moment. ³² Similarly, when Justice Stevens focuses on the moment in the video in Scott v. Harris where the plaintiff slows and waits before passing into the other lane, to countervail the “out-of-control driver” reading of the video put forward by the majority; or points out our lack of direct information establishing that the plaintiff ran red lights, he is engaging in an effort to reframe an interpretively ambiguous video by focusing on specific instants and how they reveal critical limitations to the knowledge the videos can provide. ³³

To be sure, the tools at an attorney’s disposal for ambiguous images or jigsaw images are, in essence, the same tools always in an attorney’s arsenal when engaged in the interpretation of evidence more generally. An attorney may focus attention on precisely what is (or is not) depicted in the image; focus attention on a narrower or broader context; look at how a given interpretation of an ambiguity coheres (or does not cohere) with other evidence, visual and otherwise; frame the common-sense plausibility (or lack thereof) of any given inference, and so on. Note, however, that here too, enterprising attorneys may draw on notions of expert and professional ways of seeing to challenge an otherwise-common sense reading. Charles Goodwin’s analysis of the Rodney King case, for example, emphasizes the critical dimension of offering not just a reframing of the video, but a reframing that situates it within a particular profession, and hence requires professional expertise by police “insiders” in order to be properly understood. The video was “radically transform[ed]” by the defense by being placed “within an expert frame of reference,” that re-coded and re-classified its events. ³⁴ The defense argued (successfully) that only with the assistance of police experts as guides to what was shown could the jury appropriately understand the videotape. In other words, what appeared, at first to be a fairly legible video, quite comprehensible to a lay viewer, was made, by the defense, into something far less legible than it initially appeared, transformed into a set of images requiring expert assistance for genuine understanding – it was transformed in other words, into an exemplar of the next category of image in our taxonomy.

Images requiring expertise: With our next category, evidence semi-legible to a layperson but readily interpretable to an expert, questions about how to think about the relationship between the visual evidence, the expert, and the lay interpreter almost inevitably arise. Ought we to conceive of the expert as a teacher, educating the lay interpreter to “read” the image for him or herself? Or is the layperson deferring to the expert’s authority vis-à-vis the image? ³⁵ If we believe that education of the lay juror to read the visual evidence is either impractical or epistemically impossible, should the jury still see the visual evidence at all? Is there a risk that the factfinder will be unduly persuaded by the semi-legible visual evidence, coupled with the expert’s interpretation – that the lay viewer will believe that he is now “seeing” for himself, when in fact she is simply seeing the expert’s argument in visual form? This rhetorical strategy – turning the layperson into a “virtual witness” ³⁶ – may be more powerfully effective than is legitimately warranted by the visual evidence itself.

Imagine, for example, an x-ray introduced into court showing what the expert says is an abnormal mass indicating a particular kind of tumor. “Aha,” says the layperson/juror, “now I see it.” But precisely what do they see? They have had their attention directed in a particular way by the expert, and now believe they “know” which part of the image carries special relevance. They see an indication on the visual image that the expert has told them (a) indicates a mass; (b) is abnormal; and (c) establishes a particular medical condition. It is unlikely that the expert has shown them an array of other chest x-rays for comparison, normal or abnormal, in the course of testimony (and even in the unlikely event that the expert did so, the viewer would still be largely taking the expert’s word for it about what these other images showed and meant). So: the lay observer actually has no personal knowledge that the visual indicator pointed to by the expert is abnormal; nor does she know whether or not it actually indicates a mass as opposed to having some other likely cause; nor does she have any knowledge apart from the expert’s say-so that the abnormal mass, if such it is, indicates a specific medical condition. The lay observer, to the extent that she now “knows” any of these facts, knows them as a result of deferring to the expert’s authority and believing the expert’s interpretation of the evidence. And yet, the observer may well feel as if she has seen the tumor for herself, that she “knows” of its existence in a far more direct and powerful way than if the expert had simply told her of its existence, and described it using only words.

Similarly, when a fingerprint expert accompanies her testimony with enlarged charts showing the latent fingerprint found at the crime scene and the defendant’s source print, she usually annotates and marks the images to show the array of similarities of minutiae she has found. In doing so, she directs the lay viewer’s attention toward specific markings within the images, and thereby defines in her presentation what matters, what is worthy of attention, what “counts.” If the expert is effective, the lay viewer may come away believing that she now literally sees the match for herself, and even that she now “knows” firsthand that the defendant left the print found at the scene of the crime. Parsed more carefully, she knows no such thing, and her belief to the contrary is necessarily grounded in substantial ongoing deference to the expert rather than direct personal knowledge. How many minutiae are necessary to decide that two prints match? The images standing alone cannot tell the juror that. How should any apparent visual difference between the prints be understood or interpreted? The lay examiner cannot answer that either by reference to the images. Even if the two prints seem visually similar in many particulars, what does that mean? Does it mean they share a common source, or might two fingers from different individuals nonetheless have that much visual similarity? The images of the prints, by themselves, cannot offer answers to any of these critical questions, and yet in order to know that the prints were left by the same finger, all must be answered – so if the layperson believes he or she now knows that fact, it is because he or she believes the expert, rather than because the images themselves establish it.

Thus, consciously or not, the expert may use the semi-legible display both to persuade, and also to shore up her own authority – though, to be sure, it does so in a complicated way, and often incompletely. On the one hand, the expert’s use of the visual display permits her not only to tell but also to show – and potentially to “perform” her own expertise by masterfully interpreting the visual evidence and thereby illustrating her authority to the jury. On the other hand, the expert also risks making the visual evidence seem like the “real” evidence, and the interpretation offered by the expert merely a gloss on this evidence, thereby locating the power of the evidence in the visual object as much as, or perhaps even more than, in the expert’s knowledge. This dynamic potentially generates a risk that the factfinder could develop what, from the expert’s perspective, would be too much hubris: having been taught by the expert how to interpret the semi-legible visual evidence, it might now attempt its own interpretation, unmoored from the expert’s guidance.

Does it matter if the jury believes it now understands the x-ray for itself, or “sees” the fingerprint match, or the subdural hematoma, or the letters on the license plate, and loses sight of the chains of necessary inference and deference to the expert’s authority that continue to undergird these conclusions? To the extent that the expert’s interpretation is indeed worthy of deference, we may not be acutely concerned about these vividness effects and the potentially high persuasive power of these semi-legible forms of evidence. But when these images are only semi-legible even to the expert, or when the expert to any degree overstates the legitimate interpretation of the visual evidence upon which her opinion relies, we must recognize that judges and juries may be poorly equipped to recognize this problem, or to adjust their own interpretations as a result. Take Shaken Baby Syndrome, for example, the subject of the Middleton case in the initial series of vignettes. A revisionist medical causation story has begun to take hold; some now suggest that evidence of the sort presented in Middleton – that in the absence of obvious external trauma, subdural hematomas and brain swelling visible on brain scans like CT scans and MRIs, are almost certainly caused by violent, non-accidental shaking – is not nearly as certain as many physicians have regularly claimed in court. ³⁷ Yet numerous juries have taken the physicians’ interpretation of the meaning of these visible signs as authoritative, and the ability of the jury, often, to gaze on the scans and “see” the harm allegedly done by shaking likely substantially enhances the plausibility of the physician’s claims about these injuries’ cause. Both juries and judges – and even experts themselves – may thus take the visual evidence to show far more than it actually does. ³⁸ They may erroneously conflate the statistical and the particular (e.g., believe that because images of this sort are, on average, associated with the existence of a condition, then this plaintiff now definitively has the condition). ³⁹ The expert may be able to engage in interpretive sleight of hand, with the visual image as an accessory, and the incredulous jury may simply not know enough to see the overstatements, the missing arguments, or alternative plausible interpretations. At the same time, these visual displays may have a vividness and salience that is not altogether contained by the expert’s interpretations. One risk is that the jury will too willingly accept the expert’s reading and believe it now “knows” for itself. The opposite risk, however, surely exists as well, that the jury may use the visual evidence, in conjunction with the expert’s limited teachings, in the service of “readings” of the semi-legible visual evidence that any legitimate expert would find ridiculous or fanciful.

To some limited extent, the adversarial process may offer a potential corrective to these excesses: perhaps the opposing party will make use of its own semi-legible visual evidence, or effectively challenge the interpretation offered of those visual displays that have been introduced (with or without its own expert interpretations). But in many areas, especially on the criminal side, the introduction of counter-experts is the exception rather than the norm; moreover, the adversarial process and the system of party-controlled expert witnesses certainly does not generate the appropriate incentives for testifying experts to engage in balanced, even-handed interpretation of the visual displays they use. ⁴⁰ Additionally, even when there are experts on both sides, that leaves the jury in the position of assessing competing expert interpretations for which, given the semi-legibility of the visual evidence at issue, they almost by definition lack epistemic competence. ⁴¹

Partly because of their semi-legibility, all of these forms of visual evidence may also raise particular concerns about cognitive bias. Psychologists have identified numerous ways in which our perception and interpretation can, notwithstanding our honest desire to see clearly and interpret accurately, be significantly affected by context. They have found substantial evidence of observer effects – the tendency to see what we expect to see – and hindsight bias – the belief that we have always seen or known something once we have come to know it. Vividness can also affect perceived persuasiveness more than it ought to. ⁴² Moreover, ambiguity – and semi-legibility – appears to increase our susceptibility to these cognitive biases. ⁴³ While there has been growing attention to cognitive bias in the forensic sciences, as well as in more traditional fields like radiography, courts have not yet much explored the potential impact of these biases on the interpretation of visual evidence. ⁴⁴

One final issue deserves mention. Do we always want to increase the legibility of images if we are able to? Or, more precisely, is a more legible image also a more persuasive one? It might seem that the legibility of visual evidence and its persuasive power would go hand in hand: that all else being equal, the more legible a piece of visual evidence, the more persuasive it would be in establishing that which it shows. Surely this is often the case: if the surveillance camera shows me the perpetrator’s face clearly, I will presumably feel more equipped to decide whether the defendant is the person shown in the video than if I view a blurry and distorted image. But this relation between legibility and persuasive power is not always so simple. Why so? First, a semi-legible image may make some relevant dimension of the image more salient than it would have been if the image were more legible. In Scott v. Harris, for example, the semi-legibility of the surroundings, the fact that we glimpse only blurrily the buildings and cars at the side of the road, strongly contributes to the viewer’s sensation of speed. Second, our film-generated genre conventions and interpretive norms may sometimes mean that semi-legibility helps to convey a particular mood or tone, and this may contribute to the persuasive power of the images. A shaky, blurry surveillance image captured in low light may make the scene depicted feel scarier and less secure than the same image taken in conditions generating a more legible image. If the defense in an excessive force civil suit were introducing this video to establish the reasonableness of the police officer’s decision to fire upon an unarmed suspect, this video might be significantly more persuasive than a more legible version. Third, working to interpret something, rather than simply seeing it without effort, may increase the persuasive power of that interpretation. When a viewer struggles to make sense of something – invests labor, energy, mental effort – he or she may believe in the result more deeply than if that interpretive energy did not need to be invested in the first place. An image that someone needs to be “taught” to see – like the x-ray, the brain scan, and the fingerprint, or even the interpretively ambiguous image like the Rodney King video that is parsed and reframed with expert help – may well seem to the lay viewer to be very secure knowledge indeed by the time the expert has finished her testimony.

While this foray into semi-legible questions has raised as many questions as it has answered, its main purpose has been to suggest that it is indeed a category worth attending to. More generally, I hope to have provided a helpful analytic taxonomy of semi-legible evidence, and to have shown a variety of ways in which parties might endeavor to make semi-legible evidence more interpretable to a lay factfinder. I hope also to have suggested that the connections between visual evidence and expert evidence are both deeper and subtler than we often recognize them to be. Semi-legible images cannot be said simply to speak for themselves; they must be made to speak, through the exertion of effort, expertise, or both. A superficially legible image – like the videotape in Scott v. Harris, or the video in the Rodney King case, can, through both effort and expertise, be made to seem significantly less legible than it first appeared. Conversely, an image that initially appears nearly illegible can be made, through effort and expertise, to seem far more interpretively meaningful than it first appeared, as with the images and license plate in the Bellfield case, or the claimed link between what the CT scans show and shaking.

We thus see the way in which semi-legible visual evidence creates the possibilities for a range of interpretations, all of which are affected by methods for enhancing (or reducing) the apparent legibility of the depiction. These images cannot mean absolutely anything at all – the visual depiction, notwithstanding its only partial legibility, constrains the range of the plausible – but neither can it be said to speak for itself, to be self-interpreting, or to be beyond all doubt or potential counter-interpretation. Scalia’s certainty that the video speak for itself in Scott v. Harris is misguided, but so is any approach to visual evidence that fails to recognize that interpretive plausibility is dramatically constrained by what is depicted. The existence of multiple plausible interpretations is not the same as claim of infinite, or unbounded interpretation; even the deployment of significant resources, through forms of interpretive effort and expertise cannot make an item of visual evidence seem to show whatever a party wishes it to. There are, alas, no magic formulas for assessing or delimiting the range of what is plausible, or for determining when a semi-legible image will be more or less persuasive to a viewer. But by attending to what makes an image more or less legible, the strategies that can enhance (or diminish) legibility, and the significant links between the deployment of expertise and the assessment of visual evidence, we can continue to pursue an engagement with both the messiness and the power of semi-legible visual evidence.