Phase congruency: a low-level image invariant.

摘要: Phase congruency is a low-level invariant property of image features. Interest in invariants has been limited. This surprising, considering the fundamental importance being able to obtain reliable results from operations order successfully perform any higher level operations. However, an impediment use phase detect features its sensitivity noise. paper extends theory behind calculation number ways. An effective method noise compensation presented that only assumes power spectrum approximately constant. Problems with localization are addressed by introducing new, more sensitive measure congruency. The existing developed for 1D signals extended allow 2D images. Finally, it argued high-pass filtering should be used information at different scales. With this approach, choice scale affects relative significance without degrading their localization.