Investigation Of Fast And Rotation-Invariant Image Matching Algorithm Based On Feature

The image matching algorithm is one of the cores of image matching assistancenavigation system.The capability of navigation system is mainly laid on the capabilityof image matching algorithm, which is really difficult for ideal aircraft navigation andis paid lots of attention by many researchers in the world. In order to enhancenavigation's real-time and accuracy, investigating the image matching algorithm foraircraft navigation has important significance both in theory and practice.Because the characteristic, the distance, the time and the view point of thereference image and real-time image for navigation are different, the gray and thegeometrical characteristic may change between the two kinds of images.Theconventional gray-based image matching algorithms can not meet the need of thenavigation. The real-time image may have arbitrary rotation compared with thereference image and matching must be achieved instantly, investigation of theanti-rotation matching algorithm, which has better accuracy, fast speed and strongadaptability, has realistic significance. As above-mentioned, this paper proposes a fastfeature-based method, which is a rotation invariant image matching method.Feature extraction, image matching method handling translation, image matchingmethod handling translation and rotation and fast searching strategy are discussed inthis paper.1．Feature extraction. The corner and edge are respectively extracted in thispaper, and preprocessing for images is made before the feature extraction.Results show that the performance of feature extraction with preprocessing isbetter than the performance of feature extraction with no preprocessing.2．Matching method handling only translation. At first, normalized productcorrelation algorithm and sequential similarity detection algorithm aredescribed. Then several feature-based matching methods are introduced.Finally, large numbers of experiments are made for image matching withabove methods.And the method which has best accuracy and appropriatecalculation time is chose as the similarity measurement criterion of thefeature-based and rotation-invariant method proposed in this paper.3．Matching method handling translation and rotation. At first,momentinvariants matching algorithm and circular projection matching algorithm aredescribed. Then a feature-based and rotation-invariant algorithm for imagematching is proposed in this paper. The rotation angle is estimated by thehistogram of difference of the orientation angle of the feature points betweenthe reference image and real-time image. Then the real-time image is rotatedaccording to the estimated angle,the similarity measurement value iscalculated and the position,which has the best similarity measurement value,is as the best matching position. At the same time, the rotation angle is got.4．Fast searching strategy.The improved simulated annealing algorithm is usedduring searching the matching position in order to increase the matchingefficiency.5．A great deal of experiments are made using the proposed algorithm combined with the improved simulated annealing algorithm, and its performance iscompared with the classical anti-rotation algorithms at the aspect of matchingaccuracy, matching speed and the capability of estimating the rotation angle.Results show that the proposed algorithm has better accuracy and capabilityof estimating rotation angle.