Module image 

Image XObjects and inline images → premultiplied RGBA pixmaps (ISO 32000-1, 8.9): every bit depth, the colour spaces of super::color, /Decode ranges, stencil masks painted in the fill colour, /SMask soft masks, stencil and colour-key /Masks. The sample data comes through the raster’s filter chain, its libjpeg-exact JPEG decoder and its CCITT decoder; JPX and JBIG2 have no decoder here and come out as a mid-gray block (the picture the layout model should see is “a picture”).

Loading (load: filters, codec, colour space, alpha plane) is separate from rasterizing (rasterize: the RGBA pixmap for one destination size) so a document-wide cache can decode a scan once and blit it at the pipeline’s several scales. Like docling-parse (build_bitmap_image), an image several times larger than its destination is box-averaged down by an integer factor before the blit, so a 5000-pixel-wide scan drawn 600 pixels wide costs 600 columns.

Structs

DecodedImage

LoadedImage

An image decoded to samples, with everything the rasterizer needs.

Functions

codec_reduction_shift

docling-parse’s codec_reduction_shift: how many halvings a JPEG may be decoded at (libjpeg’s 1/2, 1/4, 1/8 reduced inverse DCT) and still hold at least the pixels it is drawn onto at target_pixels_per_unit — the renderer’s bitmap_target_pixels_per_unit, docling’s render_scale of 1.0 — so the rasterizer keeps minifying and never magnifies samples it once had. A 300 dpi scan drawn full-page decodes at a quarter for the 72 dpi hint and is blitted up onto the scale-2 canvas; matching that is what lines the Rust render of a scan up with the shim’s. Capped at three halvings and never below 64 samples a side (a thumbnail is not worth the resampling risk); drawn_*_units is the axis-aligned extent of the drawn quad in PDF units.

declared_size

Decode stream to samples: the filter chain and codec, the colour space, the alpha plane. The /Width × /Height an image dictionary declares (the codec’s own size wins once decoded; this is what the reduction rule is asked with).

decode

load then rasterize.

load

Decode an image XObject (or inline image) into samples. reduction_shift asks the JPEG decoder for a 1 / 2^shift reduced decode when the image allows one (codec_reduction_shift, [may_reduce_decode]); the loaded image then reports the reduced size.

rasterize

Rasterize a loaded image as it would be painted with fill (for stencil masks), reduced so that its whole extent covers about target device pixels (None = full size).