Variational pattern analysis using Gabor wavelets

Measuring local frequencies is an effective approach to analyzing globally nonstationary, yet locally coherent images. The local frequencies may be manifested as, e.g., flowlike, granular, or oriented patterns. It is necessary to make spectral measurements accurate in both frequency and space, conflicting requirements constrained by a generalized uncertainty relationship. Such spectral measurements can be obtained from multiple wavelet-like channel filters that yield a locus of possible frequency solutions. Thus locus of solutions is maximally localized in both space/frequency if the channel filters are Gabor wavelets. A solution is obtained by imposing stabilizing terms developing naturally from assumptions on the signal. The frequency measurement problem is cast as a regularized extremum problem, leading to an iterative constraint propagation algorithm. As a to an iterative constraint propagation algorithm. As a byproduct local image contrast measurement is also obtained