Robust discretization, with an application to graphical passwords

When data or the processing on the data has some uncertainty, discretization of those data can lead to significantly different output. For example, in certain graphical password schemes, a slight uncertainty in the clicking places can produce a different password; another example is digital watermarking, where a slight change in the features can produce a different watermark. We present a discretization method which is tolerant to uncertainties (described by an error distance r, which can be adjusted to the applications). This enables us to implement very flexible graphical password schemes, and digital watermarking. (Abstract still to be completed.)