Title: A spectral variational approach for image analysis
Abstract: Linear eigenfunction decomposition is a very powerful technique used broadly in the signal and image processing community. In this talk an alternative variational processing methodology is introduced which is based on nonlinear eigenfunction analysis, referred to as a spectral approach.
Some motivation will be given and analogues to Fourier filtering methods and linear eigenfunction decomposition will be drawn. A spectral total-variation framework will be presented. The more general case of one-homogeneous functionals will be discussed as well. Applications for image representation, decomposition and texture analysis will illustrate possible benefits of this framework.