It is well known that under a variety of conditions Fourier phase is s
ufficient for image representation. The application of present techniq
ues to image reconstruction from global (Fourier) phase is, however, r
ather limited in practice due to the computational complexity. We pres
ent a new approach to image representation using partial information d
efined by the localized phase. Our scheme is implemented using the sho
rt-time (short-distance) Fourier transform. This is a generalization o
f the Gabor scheme which is well established with regard to biological
representation of visual information at the level of the visual corte
x. Similarly to processing in vision, the dc component is first extrac
ted from the signal and treated separately. Computational results and
theoretical analysis indicate that image reconstruction from the local
ized phase representation is more efficient than its reconstruction fr
om the global phase representation in that the number of required comp
uter operations is reduced and the rate of convergence is improved. It
is also implementable with fast algorithms using highly parallel arch
itecture.