`deskew`

Automatically detect and correct document skew

Usage

pdftl <input> deskew [<page_specs>...] [max_skew=<val>] [dpi=<val>] output <file>

The deskew operation automatically detects rotational misalignment (skew) commonly found in scanned documents and aligns the text horizontally.

This operation runs a page-level hybrid vector/raster deskew:

The page is rendered at a low resolution to analyze its layout.
A projection profile variance analysis (Radon-style line search) detects the dominant text angle within the specified max_skew range.
If the active text region or page size is too small for a high-fidelity line search, the page is dynamically re-rendered at a higher resolution (up to 600 DPI) to prevent coarse-grain estimation and boundary errors.
The entire page coordinate space is rotated around its visual center to correct the alignment.
Vector content, high-resolution image assets, and text rendering remains lossless because the coordinate grid itself is rotated.
Clickable hyperlinks, highlights, and annotations are transformed automatically to match their new visual locations.

max_skew=<val> (default: 10.0) — The maximum search limit for skew detection, in degrees. Restricting this window speeds up analysis and avoids false-positive rotations on non-standard layouts.
dpi=<val> (default: 75) — Resolution of the rasterized page used for skew analysis. 75 DPI is generally optimal for fast, robust line detection.
max_render_dpi=<val> (default: 600) — Ceiling on the resolution used when a small text region is re-rendered for closer analysis. This controls how much detail is captured; it is independent of coarse_res/fine_res below, which control how much of that captured detail is analyzed.
coarse_res=<val> (default: 300) — Target pixel footprint (long edge) used for the initial coarse angle search. Lower values are faster but risk missing the correct angle’s neighborhood on very fine or sparse text.
fine_res=<val> (default: 600) — Target pixel footprint (long edge) used for the fine, sub-degree angle search. Higher values improve precision (reducing residual jitter of a few tenths of a degree) at the cost of slower analysis; lower values trade some precision for speed.

Automatically detect and correct skew on all pages.

pdftl in.pdf deskew output out.pdf

Automatically deskew only the even pages, limiting search to 5 degrees.

pdftl in.pdf deskew even max_skew=5 output out.pdf

Deskew pages 1 to 3 with a higher analysis resolution of 150 DPI.

pdftl in.pdf deskew 1-3 dpi=150 output out.pdf

Prioritize speed over precision by using smaller search footprints.

pdftl in.pdf deskew coarse_res=150 fine_res=300 output out.pdf

Tags: content_modification, geometry

Source: pdftl.operations.deskew

Type: Operation