期刊名称：International Journal of Advanced Research In Computer Science and Software Engineering
印刷版ISSN：2277-6451
电子版ISSN：2277-128X
出版年度：2012
卷号：2
期号：4
出版社：S.S. Mishra
摘要：In this wo rk we p resent an overview of ima ge alignment, describ ing most of the a lgorithms and their extensions in a co nsistent framework. We concentra te on the inverse compositional a lgori thm, an effic ie nt algo rithm that we recentl y proposed. We cover the quantit y approx imat ed, t he warp up date rul e, and the g radient d escent a ppro ximation. In thi s wo rk, we p ro p o se the use of a modified version of t he correlatio n co effi cient as a perfo rm a nce cri terio n fo r the image alig nment p ro blem. The p rop osed mo difi catio n has the desi rable cha ra cteri stic of being i nvariant wit h re sp ect to ph oto met ri c di st ortio ns. Since the resulti ng simi larity m ea su re is a no nlinear function of the warp pa ra m eter s, we d evelo p t wo i tera tive sch e me s fo r its max i mi zatio n, o ne b a se d on the fo rward a dditi ve a pp r oa ch and t he seco n d o n the in verse co mp osit i o na l met ho d . As is cust o ma ry in it erati ve op timi zatio n, in e a ch iteration, the no nli nea r ob jecti ve function is app ro xi mat ed by an a lternativ e e xp re ssio n for which the co rrespo n di ng op timi zatio n is simple. In our ca se, we prop o se an effi cient ap prox i ma tion tha t lea ds to a clo sed -form so lutio n (p er it erat ion) whi ch is of low com p ut atio nal comp lexi ty, t he latter pro perty being part icula rly strong in our inverse versio n. The p ro p o sed sc hem es are t est e d a g ai nst the Forward Ad di tiv e Luca s-Ka na d e a nd t he Si mu lta n eo us In verse Co mp ositi o nal (SIC) algo rithm throug h simul ati o ns. Und er noisy conditions a nd p ho to met ri c di stort io ns, our fo rwa rd versi on a chie ves mo re a ccurat e alig nme nts and ex hibit s fa ster co n verg enc e, whe rea s our i n verse versio n ha s similar per fo r ma n ce as the SIC a lgorithm but at a lower comp uta tio na l co mpl exit y.
关键词：Ima ge a lignment; Lucas-Kanade; a unifying framewo rk; a dditive vs. compositio nal algo rithms; for- wards vs. ;inverse alg orithms; the inverse composit iona l al gorit hm; effi ciency; steepest d escent; Gauss-Newto n; Newto n; Levenberg -;Marquardt .
