Robust Image Signatures

Whatever the task, visual content needs to be represented somehow. How can we create visual signatures that are robust to all different types of media (documents, high quality photographs, medical scans, video streams) and a wide range of application scenarios? Research is conducted into developing flexible image signatures that are robust as possible to variations in task and data set, but are adaptive to different scenarios and requirements, without the need to custom-design a representation for each new task; but can be tailored to an application if the end application requires it.