To resolve the mismatching problem in image registration and image mosaic

this paper proposes an improved feature matching principle based on vector similarity.First

by detecting feature points in scale space

SIFT (Scale Invariant Feature Transformation) vectors

which represent local properties

are computed;the feature points are matched by using vector similarity method and mismatching point couples are further deleted with the application of mutual mapping theory.And then

the transform matrix is calculated by random sample consensus algorithm.Furthermore

it is optimized by Lenvenberg-Marquardt (L-M) algorithm.Finally

through image fusion

image mosaic is realized.The experimental results indicate new principle improves the matching precision.The method can deal with image registration and mosaic with projective transformation.