How does browser extraction preserve original PDF image quality?
Browser extraction pulls embedded raster streams directly from the PDF’s indirect object table inside a Web Worker, copying JPEG bitstreams verbatim without rendering pages to an HTML canvas. Because 0 bytes leave your device and no canvas resampling occurs, every extracted photo keeps its exact native pixel dimensions and compression history.
Saving DCTDecode photos verbatim without canvas re-encoding
Most PDF generators embed photographs as /DCTDecode streams, which are complete JPEG byte sequences wrapped inside a PDF dictionary. According to the Library of Congress JPEG Encoding Format Description, re-saving a JPEG through a browser canvas introduces a second round of lossy DCT quantization.
Instead, this extractor writes the raw stream bytes straight to .jpg (and .jp2 for JPXDecode streams). If you later edit those photos and want to reassemble a clean packet, you can combine extracted pictures back into a PDF without leaving your browser.
Decoding FlateDecode streams and SMask transparency into PNG
Screenshots, logos, and diagrams stored with FlateDecode, LZWDecode, RunLengthDecode, ASCIIHexDecode, or ASCII85Decode are decoded in memory without canvas interpolation. When a stream uses PNG row predictors (Predictor >= 10), the worker reverses the predictor filter row by row before encoding.
If an image carries a matching 8-bit SMask soft-mask stream, the extractor merges both buffers into an RGBA (colorType 6) or gray-plus-alpha (colorType 4) PNG. Standalone ImageMask bitmaps and mask streams are excluded so transparency masks do not appear as duplicate black-and-white files.
Filtering icons with the minimum-side pixel threshold
PDF exports from slide decks and web pages often embed dozens of tiny bullets, border tiles, and UI badges alongside full-size photographs. Filtering by pixel dimensions before packaging keeps your output folder clean and focused on usable assets.
Choosing between 1 px, 64 px, and 200 px thresholds
The Skip images smaller than control checks both width and height against your chosen threshold before decoding:
- Keep everything (1 px): Extracts every non-mask raster XObject in the file, including tiny icons, spacer tiles, and bullet bitmaps.
- 64 px (default): Skips small icons and decorative bullets while keeping standard thumbnails, inline figures, and content graphics.
- 200 px: Restricts extraction to large photographs, scanned pages, and high-resolution figures.
Handling color spaces, bit depths, and ZIP packaging
The decoder handles 1, 2, 4, and 8 bits per component across DeviceGray, CalGray, DeviceRGB, CalRGB, DeviceCMYK, ICCBased (1, 3, or 4 channels), and Indexed palette color spaces. CMYK samples are converted into RGB channels before writing chunks compliant with the W3C Portable Network Graphics (PNG) Specification.
Files are named by their first page appearance (page-1-image-1.jpg), or image-1.png for unreferenced objects. When more than one image is extracted, they are packaged into a single ZIP archive; after retouching assets, you can package edited PNG graphics into a document in one step.
Image stream formats and extraction behavior
How a PDF stores an image determines both the output file extension and whether any pixel conversion occurs. The table below summarizes how each PDF stream type is handled by the worker.
| PDF stream filter or type | Output format | Quality and alpha handling |
|---|---|---|
DCTDecode (JPEG) |
.jpg |
Byte-for-byte copy with zero re-encoding |
JPXDecode (JPEG 2000) |
.jp2 |
Byte-for-byte copy of embedded codestream |
FlateDecode / LZWDecode |
.png |
Lossless 8-bit PNG (zlib level 6) |
FlateDecode + 8-bit SMask |
.png |
Merged RGBA or gray+alpha transparency |
CCITTFaxDecode / JBIG2Decode |
Skipped | Reported in skipped count on result panel |
| Vector paths and text | Not extracted | Stored as drawing operators, not XObjects |
Every Image XObject lives in the PDF structure defined by the PDF Association ISO 32000-2 Specification. Checking /Subtype /Image across indirect objects ensures embedded images inside nested /Form XObjects are also discovered and named by page. When a page contains vector artwork instead of raster XObjects, switch to our lossless PDF to PNG converter to rasterize complete pages cleanly.
When to extract embedded images versus render whole pages
Extracting embedded objects is ideal when you want raw source photos without surrounding margins, headers, or text overlays. However, some documents do not store their visuals as raster Image XObjects at all.
Handling vector diagrams, fax scans, and skipped streams
CAD blueprints, Illustrator charts, and LaTeX figures are drawn with vector paths rather than raster pixels, so they will not appear in an Image XObject scan. Similarly, older monochrome fax scans encoded with CCITTFaxDecode or streams with custom Decode arrays are counted as skipped.
Tip: If your PDF contains vector logos or fax-compressed scans that show up as skipped or missing, switch to a full-page rasterizer to capture the entire rendered page cleanly.
Converting full pages when artwork is not a raster XObject
When you need a crisp image of a vector diagram, chart, or multi-layer collage exactly as it appears on the page, full-page rendering captures every layered element in one flat frame. For multi-page photographic PDFs where smaller file sizes matter more than lossless edges, you can render full PDF pages as JPEG files in your browser.
Lossless PNG encoding inside the extractor compresses pixel rows using the IETF RFC 1951 DEFLATE Compressed Data Format at compression level 6, keeping file sizes compact without discarding a single pixel value. Each extracted file preserves the source width and height in pixels so graphics drop straight into design or documentation workflows.