Guide

Clean a Noisy Image Before Vectorizing

Separate unwanted noise from intentional marks. Use reversible edits and compare the original with each prepared candidate before deciding whether to trace it.

Free Protected Preview

Protected vector preview

No account is required to create or view the protected preview. The preview accepts one PNG, JPEG, or WebP file up to 4 MiB; other source formats must be exported or converted to one of those formats first. Editable SVG export requires paid or complimentary access. Review the exported SVG in the appropriate downstream software; non-SVG formats and machine instructions require separate conversion.

Open Protected vector preview

Locate the cause before filtering

Noise diagnosis and a conservative response

Observation
Dust isolated on plain paper
Try on a copy
Remove only confirmed dust with local retouching.
Check for damage
A period, dot or pencil accent may be intentional.
Observation
Paper grain across the drawing
Try on a copy
Improve capture or selectively adjust the background.
Check for damage
Faint lines can disappear with the paper texture.
Observation
JPEG blocks around edges
Try on a copy
Seek a less-compressed original; compare mild denoising if unavailable.
Check for damage
Softened corners and merged gaps may cost more than the removed blocks.
Observation
Uneven shadow or glare
Try on a copy
Retake the photograph with even illumination.
Check for damage
A single threshold may erase one side while leaving the other noisy.
Noise diagnosis and a conservative response
ObservationTry on a copyCheck for damage
Dust isolated on plain paperRemove only confirmed dust with local retouching.A period, dot or pencil accent may be intentional.
Paper grain across the drawingImprove capture or selectively adjust the background.Faint lines can disappear with the paper texture.
JPEG blocks around edgesSeek a less-compressed original; compare mild denoising if unavailable.Softened corners and merged gaps may cost more than the removed blocks.
Uneven shadow or glareRetake the photograph with even illumination.A single threshold may erase one side while leaving the other noisy.

Compare levels, threshold and blur

A tonal or levels adjustment can separate paper from ink while retaining intermediate shades. Thresholding collapses shades into two classes: it may suit a deliberately binary mark but destroys tonal shading. Neither method knows which gray marks are part of the drawing. Keep a separate candidate rather than replacing the capture.

Mild blur or denoising is an alternative to test for compression noise, not a required first step. Blur can soften both noise and deliberate lines. Use a small crop containing the narrowest gap and faintest stroke to compare before and after; undo if those features merge or vanish. No fixed radius works for every source size. Exact GIMP or Photoshop menus depend on the installed version and are not prescribed here.

Worked diagnostic: the speck and the dot

Illustrative example: a scanned wordmark has dust next to the dot of an i. Removing every small island would remove both. Check the original drawing or a cleaner reference, retouch only the dust, and preserve the dot. A trace may combine or omit small regions; it is not true that every speck becomes a separate path.

Trace only after checking the prepared source

  • Record the original, the edited copy and the adjustments you applied. View them at the same size.
  • Check thin strokes, corners, holes and intentional texture before uploading. A larger pixel count does not supply missing edge information.
  • Create a protected preview and compare the same problem locations. Authorized SVG export permits later geometry inspection; it does not certify cleanup.
  • If noise is still substantial, seek a better capture or repair selected geometry in an editor. Repeated filtering and resampling can compound damage.

Owned preparation and actual trace comparisons

Original: authored dust and intentional dot.
Original: authored dust and intentional dot.
Actual Fixed V2 trace of this preparation; inspect against its input.
Actual Fixed V2 trace of this preparation; inspect against its input.
Delete only known dust; preserve the intentional dot.
Delete only known dust; preserve the intentional dot.
Actual Fixed V2 trace of this preparation; inspect against its input.
Actual Fixed V2 trace of this preparation; inspect against its input.
Gaussian blur sigma1; fine detail may change.
Gaussian blur sigma1; fine detail may change.
Actual Fixed V2 trace of this preparation; inspect against its input.
Actual Fixed V2 trace of this preparation; inspect against its input.
Threshold160; tonal distinctions are deliberately discarded.
Threshold160; tonal distinctions are deliberately discarded.
Actual Fixed V2 trace of this preparation; inspect against its input.
Actual Fixed V2 trace of this preparation; inspect against its input.

All four160 × 100 inputs are owned digital fixtures. Dust is the1-pixel-radius mark at(130,24); the intentional dot is radius3 at(85,23). The retouched input removes only the authored dust; blur and threshold are separate Sharp operations. Observed results: original and selectively retouched traces both shrink and darken the A counter relative to the input. Retouch removes known dust and retains the i-dot, but does not make that trace acceptable. Blur rounds the A tip/corners and changes color; threshold turns blue to black and yields coarser edges. Reject these outputs when those features matter; fewer paths or bytes is not a quality verdict. This is not a scan or phone capture.