XKCD's bobby tables comes to mind. Sanitize your inputs. If you point & click and you immediately process what your camera sees, there is great risk in that.
I want a photo to be a photo. If I want to scan a QR code for the purpose of scanning a QR code, I use some special app (and Ι block it's Wifi/3G connectivity to enusre the QR app will not leak what Ι just photographed).
Edit: I follow the Steve Gibson school of thought. I want the "thing" to do the "thing", and nothing but the "thing". Camera should do camera-ing (adjust camera-related-attributes). QR app should do QR-app-things (show me in clear text the QR code and ask me what do I want to do with it)(register WiFi, visit a website, etc).