geonative-image

Pure-Rust reader for image + world-file rasters — the "JPG/PNG with a 6-line .jgw/.pgw next to it" upload case.

What v0.1 covers

• JPEG (.jpg / .jpeg) decode via jpeg-decoder
• PNG decode via png
• World file sidecars (.jgw / .pgw / .wld) — 6-line ASCII format:

  0.5        ← pixel width
  0          ← rotation y
  0          ← rotation x
  -0.5       ← pixel height (typically negative for north-up)
  144.96     ← world x of centre of pixel (0, 0)
  -37.81     ← world y of centre of pixel (0, 0)
  

• CRS hint via with_crs() builder — world files don't carry a CRS, so the caller specifies it (or defaults to Crs::Unknown)
• Implements geonative_core::raster::RasterLayer — same trait that geonative-geotiff::GeoTiff implements, so geonative-convert treats them interchangeably

v0.1 scope cuts

• No pyramid — these are single-resolution images; for multi-level output you convert to COG via geonative-geotiff
• JPEG2000 (.jp2) — needs jpeg2000-decoder; deferred
• WebP — common in modern web maps; deferred
• TIFF — covered by geonative-geotiff, not here
• Writer — these formats are inputs only

Pi-friendly resource use

Single-image decode loads the full pixel array (the whole RasterTile). For a 4096×4096 RGB image that's ~50 MB peak RAM. For Pi serving, the recommended workflow is:

# 1. User uploads big_drone_ortho.jpg (multi-MB)
# 2. Convert it to COG once, off the hot path
geonative convert big_drone_ortho.jpg ortho.cog
# 3. Tile server reads from ortho.cog via mmap — Pi-friendly RAM forever

For tiny inputs (< 50 MB images) the raw image path is fine; for large inputs, convert to COG first.