Matching Algorithm Research Based On Moderate-strong Geometric Constraints

With the development of information technology and computer networks, theunderstanding of the computer for multimedia information becomes increasinglymature under the explosive growth of multimedia information. Image matching is afundamental problem in the field of computer vision, image retrieval and is also abasic problem of other areas such as pattern recognition. Image matching is dividedinto global feature-based method and local feature-based method. The second methodhas become the focus of research in this field due to its unparalleled advantages ingeometric transformation of picture such as rotation, scale changing and other realgeometric changing. The establishment of geometric constraints between features toimprove the accuracy rate of matching has become a hotspot and difficulty. Theimage matching algorithms only consider weak geometric constraints to improve theaccuracy of the match.In this paper, the SIFT feature points are used as image local features. SIFTfeature points are rotation invariant, scaling invariant, translation invariant and affineinvariant to a certain extent. The Maximally Stable Extremal Regions (MSER) arealso used as image local features. As the region based features, MSERs perform bestin comparison with other regional features. Taking the existing problems of geometricconstraint into account, this paper proposes a moderately strong geometric constraint.This method can overcomes the shortcomings of weak geometry constraint in scaling,rotation, affine attacking. Experimental results show that this method has a highaccuracy in image matching.