Decoupling Global Features by Imposing Orthogonality on the Gradients: Analysis , Synthesis and Signal Processing

Abstract

Sets of features having mutually orthogonal gradients (i.e., orthofeature sets) have remarkable properties that make them very different from statistical-based methods, and that can be exploited for a wide range of problems. So far we have demonstrated their application to data analysis (mainly texture classification and statistical regression), but lately we are exploring also its potential in signal processing and synthesis.