[ANN] GeoIPTC 1.0 : geocode your images using metadata standards (IPTC and XMP)

Question

Hello,
GeoIPTC is an innovative tool to add geographical identification to your images (geocoding) according to IPTC and XMP metadata embedded in images.
GeoIPTC reads all IPTC and IPTC Core metadata embedded in an image, isolates localisation data (City, Location, Province/State, Country Code and Country Name), then sends a request to a Geocoding Web Service and finally gets latitude/longitude coordinates.
To say it briefly: GeoIPTC gets geographical information from the postman address embedded in image.
Main features
* Gets full sets of image standard metadata
* Automatically gets longitude/latitude from widely used Geocoding Web Services like Google Maps, Yahoo! Maps or GeoNames.
* Locate automatically images on Maps Services like Google Maps, Yahoo! Maps and MapQuest
* Generate automatically KML or KMZ placemarks and locate single (one image) or multiple (many images) placemarks on Google Earth with full IPTC/IPTC Core metadata
* Set automatically single or multiple images with appropriate Exif GPS coordinates
More information and download:
http://www.geoiptc.com/EN/Index.html
With best regards
Patrick Peccatte
www.softexperience.com

Answer 1

Note the last paragraph in the original post which explains how Flickr handles resolution changes without breaking the rectangles. Perhaps a more straightforward approach would be to store both the rectangle coordinates for each rectangle AND the image's width and height at the time the rectangles were created. Then if the image is scaled, the rectangles can be scaled accordingly. For example, if a rectangle is at x=100,y=200,w=50,h=70, the "original dimensions" are 1000x2000, and the image's current dimensions are 500x1000, then an application could assume the image has been scaled down by (500/1000=) 0.50, and so the x,y,w,h of each rectangle could also be multiplied by 0.50 (x=50,y=100,w=25,h=35). I'd recommend storing floating-point values so that multiple scalings don't make the rectangles "creep".
Of course, it is possible that really the image was cropped rather than scaled, but fairly unlikely you would happen to crop it the same percent in both directions. Even if an image is cropped, someone could manually repair the rectangles by sliding and scaling them until the fit back on the faces (unless most of the faces were cropped off).
A rectangle-aware application could scale and translate the rectangles during a crop or scaling operation, but other applications could still survive scaling if they at least preserve the XMP tags as-is.