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 paper, we co ver the quantity approximated, the wa rp update rule, a nd t he g radi ent descent a ppro ximation. In future pa pers we wil l co ver t he choice of the norm, how to a llow linear appearance variation, ho w to imp ose prio rs on the para meters, and vari ous techniques to avoid l ocal minima. Since the Luca s-Ka nad e algorithm was pro posed in 1981 image alignment ha s be- co me one of the mo st widel y used t echniques in computer vi sion. Ap plica tions ra nge from opt ical flow and track ing to la yered motion, mosaic construction, and fa ce coding. Numerous alg orithms have been proposed a nd a wid e variety of ex tensions have been made to the o rigi nal formulation. We present an overview of image alig nment, d escribing most o f the algo rithms and their extensions in a consi stent framework. We concentra te on the inverse co mpositional a lgorithm, an effi cient algo rithm that we recent ly p ropo sed. We exa mine which of the extensions to Lucas-Kanade can be used wit h the inverse co m p osit io nal alg orithm witho ut a ny si g nifi c a nt loss of e ffici e nc y, a nd which ca nno t.
关键词：Image alig nment; Luca s-Ka nad e; a unifying fra mework; add itive vs. compositional alg orithms; forward s vs. inv erse ;algo rithms; t he inverse co mpo sitio nal a lgorithm; ef ficie ncy; steep est descent; Gauss-Newto n; Newto n; Levenberg -Ma rqua rdt
