Hybrid?
Why not write a client app to control the camera and then upload to the web-app for analysis and search, as it has to go online for the search anyway it's not adding /that/ much overhead as long as it keeps the image at a nice compression (though obviously not overly so as to render the analysis useless).
This way the only client apps needed would be to control the camera and the rest hooks into the web API keep differing code base development to a minimum.
Simples, as a certain meerkat would say.
Though before they go and chase Apple users, could you get the one running on my Froyo Desire to work a little better first 8)