Towards a method for parametrizing models of early vision using psychophysical data

摘要

Parametrizing computational models of vision, i.e. choosing values for their free parameters, is important both for applied computer vision and for developing the models themselves. A new, principled technique for parametrizing computational models of early vision is presented. It requires that some specific mathematical relationships are found between properties of the input to a model and some suitable psychophysical data. Comparing these either parametrizes the model, allowing it to be implemented as a computer vision system, or refutes it, motivating further modelling. The application of the technique to two computational models of early vision is demonstrated.