I would like to see the results of a bouble-blind test based on multiple negative and positive samples and at two each of least a dozen different camera phones (without extensive modification of the built-in lens, only the added sample viewer/light).

I have a feeling this wouldn't work well across-the-board without some extensive phone-specific coding, since, as mentioned above, the manufacturers use all kinds of tricks to make people think their crappy little phone takes great pictures. (other tricks include boosting saturation of greens and reds, performing noise reduction on low-saturation reds (ie skintones), and performing soft blur on blue tones, and dropping saturation of pictures taken at night and increasing contrast.

These "tricks" aren't bad per se, as they result in decent looking images from relatively low-quality and very tiny hardware... the problem is that they lead to unreliability when you wish to apply them as tools in the medical trade for detecting infectious diseases.

"I wish to extend my apologies, sir, you do not, in fact have AIDS or syphylis.... damn it, I need to buy a better phone, but my carrier has a crappy selection! Anyway, have a great day, and my apologies for quarantining you for so long! Take care now!"

 Reply