Abstract: Aiming at the problems that the existing image registrations may trigger inaccurate registrations or miss some registrations,an image local geometric registration based on improved scale invariant features was proposed．The approach improves scale invariant features,designs SI(E)FT (scale invariant edge feature transform) by constructing edge scale space and combines scale invariant feature points and scale invariant edges．Based on the improved scale invariant features,the local transforms between two images are searched to implement image local geometric registration．Experiments show that the complementary combination of SIFT points and edges provides more registration information and reduces registration errors,and the approach is insensitive to scale,noise,deformation,light,etc.,can register mo-ving objects and truly reflects image registration status．

Key words: Scale invariant edge feature transform,Scale invariant feature transform,Image transform,Local geometric registration