Sketch Image Effect

How a sketch effect is built

Nearly all sketch effects work in three stages: convert to grayscale, find the edges where brightness changes sharply, and then either darken the edges on a light background or invert the image and use the edges to build a hatching pattern. The result is a representation of the structure of a photograph rather than a drawing of its subject, which is why it works well on a building or a face and poorly on anything where colour carries the meaning.

• Step one: convert to grayscale, so colour does not confuse the edge detection.
• Step two: find edges, where brightness changes sharply.
• Step three: render the edges as lines, or use them to build hatching.
• The result represents structure, not appearance.

Strong edges are what makes it work

An edge detection algorithm looks for places where brightness changes quickly, so an image needs strong, well defined boundaries for the effect to produce a recognizable result. A building against a sky, a portrait with a defined jawline and hair against a background, a tree against the light, and a high contrast product shot all give the algorithm plenty to work with. A low contrast scene, a busy natural landscape, fog, and a flat grey wall give it almost nothing, and the result is a muddy grey field with a few random marks.

• Strong boundaries give the effect something to draw.
• Building against sky, a defined portrait, a backlit tree: these work well.
• Low contrast, fog, and a flat wall give almost no edges.
• A busy natural scene produces noise rather than recognizable lines.

NoteSketch effects need contrast. A flat, low contrast photo will look like a grey smudge.

Detail level is a trade-off

More detected detail produces a richer sketch, and it also produces a noisier one. At a high detail setting, texture in the original, such as foliage, gravel, or fabric weave, becomes a dense mass of lines that obscures the subject. At a low setting the result is clean but loses the features that make the subject identifiable. The right level depends entirely on the subject, and for a portrait it is usually much lower than for a building, because a face rendered with every pore and hair visible does not look like a drawing.

• High detail: richer, but texture becomes noise that obscures the subject.
• Low detail: clean, but may lose identifying features.
• A portrait usually needs a lower setting than a building.
• A face rendered with every hair visible does not look like a drawing.

A real drawing versus a filter

A sketch effect is a deterministic operation on the pixels, and the result has the same awkward quality as any algorithmic treatment: it looks convincingly like a drawing only at a glance, and the artefacts become visible on inspection. Textures repeat, lines follow the image noise rather than the form, and a hand would never draw the same detail twice. That is a limitation of the technique rather than a fault in the implementation, and it is worth knowing before spending time tuning an effect that cannot reach the quality of an actual drawing.

• The result is deterministic, so it looks convincing only at a glance.
• Textures repeat and lines follow image noise rather than form.
• Artefacts become visible when inspected closely.
• A real drawing has variation and intention that a filter cannot produce.

NoteIt reads as a drawing from a distance and as a filter up close. That is inherent to the technique.