A new algorithm for interpolating the missing data between two adjacen
t medical images is presented. Our method is useful for solving the in
terpolation of any region-represented images of an object to be recons
tructed, even when the object is stretched abruptly, branched or hollo
w, as often occurs in medical images, which cases can not be handled w
ell by existing methods. When this algorithm is applied, the nonoverla
pped regions of the same object in the two base images are first extra
cted and encoded by chamfer distance code on every pixel in these regi
ons. Then, the outer edges of the nonoverlapping regions are shrunk in
ward simultaneously so that the stretched edges reach the edges of the
overlapping regions at the same time. The distance codes in nonoverla
pping regions are used to limit the shrinking of these edges in the in
terpolation process. The proposed method also provides object centrali
zation and enlargement operations to obtain stable and reasonable resu
lts in complicated case. The experimental results show that the propos
ed method is more effective and efficient in resolving general interpo
lation tasks than the existing methods (S. P. Raya and J. K. Udupa, IE
EE Trans. Med. Imag. 9, 32-42, 1990; G. T. Herman ef al., IEEE Comput.
Graph. Appl. 12, 69-79, 1992; J. F. Guo et al., Comput. Med. Imag. Gr
aph. 19, 267-279, 1995). (C) 1997 Elsevier Science Ltd.