We’re excited to announce PL/IMAGE, a native PL/SQL package for decoding, editing, and re-encoding PNG and JPEG images entirely inside the Oracle database — no external Java, no OS-level image libraries, and no DIRECTORY objects to manage.
Why native image processing in PL/SQL?
In many Oracle environments — especially Oracle Cloud Infrastructure and Autonomous Database — Java stored procedures, native (C) compilation, and filesystem access simply aren’t available, or aren’t something you want to depend on. Traditionally, even a simple edit like generating a thumbnail, correcting EXIF orientation, or stitching a watermark onto a photo meant shipping bytes out to another tier and back. PL/IMAGE removes that round trip: the whole load–edit–save cycle happens in PL/SQL, working directly on BLOBs.
A simple lifecycle
Every operation follows the same shape: load an image from a BLOB, apply one or more edits, save it back out.
load_image → one or more image operations → save_image
LOAD_IMAGE returns an independent, session-local image reference; the caller may release the source BLOB right after. SAVE_IMAGE encodes the current pixels, releases that reference automatically, and hands back a fresh output BLOB the caller owns.
What PL/IMAGE can do today
Loading, saving & introspection
- LOAD_IMAGE decodes a PNG or JPEG BLOB into a fresh, independent image context.
- SAVE_IMAGE encodes the current pixels back to PNG (default) or baseline JPEG, with quality, compression level, and filter control, then releases the context.
- GET_INFO reports format, MIME type, dimensions, bit depth, color model, alpha, and any pending EXIF orientation — without altering or releasing anything.
- GET_VERSION reports the package version.
Geometry
- CROP extracts an exact pixel rectangle.
- RESIZE scales to explicit dimensions, a single dimension with proportional scaling, or a scale factor — with AUTO, NEAREST, BILINEAR, or AREA filtering.
- THUMBNAIL fits an image into a target box with CONTAIN or COVER behavior, optional upscale control, and alignment.
- ROTATE turns an image 90, 180, or 270 degrees; FLIP mirrors it horizontally, vertically, or both.
- PAD extends the canvas with a solid background; BORDER draws one inside the existing canvas.
- NORMALIZE_ORIENTATION applies and consumes pending EXIF orientation before other geometric edits run.
Color & alpha
- GRAYSCALE converts to luminance while preserving alpha exactly.
- INVERT flips every color sample; alpha is left untouched.
- THRESHOLD produces pure black-and-white output at a chosen brightness cutoff.
- FLATTEN composites the whole image over an opaque background, removing transparency — the explicit step before JPEG output.
Compositing & stitching
- OVERLAY composites one image over another with source-over blending and adjustable opacity; multiple overlays apply in call order.
- STITCH combines several images into a single canvas, horizontally or vertically, with alignment, gap, and background control.
A quick example
Loading a photo, correcting its orientation, generating a 400×400 thumbnail, and re-encoding it as JPEG takes four calls:
l_src pl_image.t_image_ref;
l_out blob;
begin
l_src := pl_image.load_image(p_source_blob);
pl_image.normalize_orientation(l_src);
pl_image.thumbnail(l_src, p_width => 400, p_height => 400);
l_out := pl_image.save_image(l_src, p_format => 'JPEG', p_jpeg_quality => 85);
end;
What’s next
The current specification covers loading, saving, geometry, color, compositing, and stitching. REMOVE_COLOR, DRAW_TEXT, and GET_METADATA are already scoped as the next additions to the API.
We’d love to hear what you’d build with a fully native, in-database image pipeline — drop your questions or use cases in the comments.
