How Are the Lines That Describe a Scene Chosen?

Remove color, texture, and shaded tones from an office photograph, and the result need not be an empty surface. A few object boundaries can still communicate how the room is arranged. Yet not every line performs the same job. The edge of a desk and a small detail on its surface may use similar amounts of ink without contributing equally to understanding the scene. For visual perception, the interesting question concerns not only how much detail a drawing contains, but which details become lines.

Research by Heping Sheng, John Wilder, and Dirk B. Walther, published in 2021, examines representations of complex real-world scenes through three experiments. Participants' drawings are compared, independent observers rank how well they represent their photographs, and selected contours are then tested for their contribution to rapid scene categorization. The design creates a measurable connection between choices made by the person drawing and information available to the person viewing. That connection is not equivalent to aesthetic taste or the value of an artwork. [1]

Making the hand's choices visible

The first experiment recruited 14 students studying visual arts or architecture and 44 psychology students. One art student and two psychology students were excluded from analysis, leaving groups of 13 and 42. The art group had greater artistic experience, but the groups also differed in age and some demographic characteristics. Group differences therefore cannot be attributed exclusively to the causal effect of art education. [1]

The task was not drawing a photograph from memory. Participants traced important contours on a semitransparent layer over the photograph on a digital tablet, while the original photograph also remained visible as a reference. The images represented beaches, city streets, forests, highways, mountains, and offices. Three photographs from each category gave a total of 18 scenes. Tracing made it easier to compare whether different people selected boundaries at corresponding locations.

Participants were instructed to avoid shading, written labels, and unnecessary repeated lines. They had up to ten minutes per photograph and could finish earlier. Psychology students completed five trials; the art group was asked to produce as many drawings as they comfortably could during two hours. Spatial coordinates and stroke timing were recorded. The study collected 194 drawings from artists and 159 from the other group. Adding 18 drawings commissioned for an earlier study created a comparison pool of 371 drawings. [1]

Representing a scene rather than winning approval

The second experiment used a separate group of 37 psychology students to rank the drawings. Sets corresponding to each photograph were printed on cards, which observers physically arranged according to how well they represented the source photograph. This measure did not ask which style the observer preferred or what a drawing might be worth. It compared alternative contour representations of the same scene against a visible reference. [1]

The researchers represented the amount of drawing detail through total pixel count. Drawings associated with greater artistic experience and drawings containing more pixels received higher rankings in this setup. This does not imply that every added line independently helps. A judgment about the whole drawing and an assessment of the information carried by individual contours answer different questions.

To address the latter question, the researchers developed a contour-matching method. It sought lines representing corresponding scene elements across drawings, despite differences in length or small spatial variations. Unmatched contours were added to a reference, building a combined set of the different contours drawn for each photograph. Comparing individual drawings with that combined reference then established how often each contour appeared. The proportion of drawings containing it became the study's consistency score.

Frequently selected and drawn early

Consistency here does not mean that a line was drawn neatly with a ruler. It describes how often different people chose to depict that contour. Most participants might select an object's boundary while fewer include a small surface detail. The measure therefore concerns agreement across people's selections, not a single person's motor precision or geometrical perfection.

The researchers also compared this agreement measure with drawing order. Among the 353 drawings with timing information, more consistent contours tended to be drawn earlier. The 18 commissioned drawings lacked timing records and were excluded from this analysis. The result was compatible with early selections prioritizing contours that more people regarded as important. It did not prove that drawing a contour earlier causes better perception; stroke order was not randomly rearranged and tested. [1]

These first stages leave a useful question open. Agreement about which lines to draw does not necessarily establish that those lines help someone recognize the scene. Shared habits could also produce shared selections. The third stage narrows that gap by transferring the selected contours into an independent perceptual task.

Similar line material, different information

Each photograph's combined contour reference was divided into two versions. One contained the most consistently selected contours; the other contained the least consistent ones. The halves shared no contours, and each contained approximately half of the reference's total line pixels. The matching concerned approximate pixel quantity, not exactly equal numbers of strokes. This made it harder to explain an advantage solely through one version containing more visible line material. [1]

The methods report that 25 psychology students participated in the third experiment. This task was added after another rapid scene-categorization experiment. Following practice and testing in that main task, participants viewed 36 half-drawings, each presented for 53 milliseconds. A mask followed the image, and the observer selected one of six scene categories with a keypress. No correctness feedback was provided during the additional test block. This was not a task allowing prolonged inspection, naming every object, or reconstructing details later from memory.

The paper reports categorization accuracy of 43 percent for low-consistency halves and 63 percent for high-consistency halves. Random selection among six categories would yield approximately 16.7 percent accuracy. The reported difference supports the conclusion that similar quantities of line pixels need not convey similar perceptual information. However, the source contains a reporting ambiguity: its methods list 25 participants, while the paired test in the results has 14 degrees of freedom. This article does not invent an explanation for that undocumented sample transition. It attributes the percentages to the authors' report rather than claiming independent verification of the exact analysis sample. [1]

The particular contribution of object boundaries

The contours were also classified by their physical origins. High-consistency halves contained more pixels belonging to occlusion or depth boundaries, typically the boundaries of objects. Low-consistency halves contained more pixels associated with changes in surface orientation and with contours that could not be clearly assigned. The study did not report the same clear differences for texture boundaries and cast-shadow edges. These comparisons suggest that different kinds of lines do not have interchangeable information functions. [1]

They do not establish a single recipe for every drawing. The research used 18 selected photographs, a particular tracing task, and brief presentation conditions. It was not a new eye-tracking or brain-imaging experiment, and the neural cause of contour agreement was not directly measured. The design nevertheless makes a valuable distinction by moving from production to independent ranking and then to rapid perception. Describing a scene with fewer lines is not merely subtraction: it involves deciding which relationships to preserve. Perceptual clarity depends on the contours' roles in the scene's structure as well as on the amount of ink.

Sources

[1] Heping Sheng, John Wilder, and Dirk B. Walther (2021), Where to draw the line?, PLOS ONE 16(11): e0258376. Open-access primary research.

Source access and factual review: 14 October 2026. Methods, results, and discussion for all three experiments were directly read. Raw data and analysis code were not rerun; the discrepancy between the third experiment's participant count and test degrees of freedom is disclosed. No illustrations were reproduced or added. Original explanation based on the Creative Commons Attribution-licensed research, with attribution to Sheng and colleagues.