Circular Arc Matching Algorithm Based On Feature Descriptor

Feature matching plays an important role in many applications,such as 3D reconstruction,object recognition,object tracking and visual navigation,and so on.Feature matching method based on feature descriptor is widely applied in all kinds of feature matching methods.The image in real life not only has point feature and line feature,but also has circular arc feature.However,compared with point matching and line matching,there are few researches on circular arc matching.Especially,the research on circular arc matching based on feature descriptor is even less.This paper is proposed based on this background.Circular arc matching algorithm based on feature descriptor involves three steps,which are circular arc extraction,circular arc description and circular arc matching.The main contents of this dissertation are as follows:Firstly,based on the idea of scale space,the scale space is constructed by Gauss Pyramid by using the method of Gauss fuzzy and down sampling.Circular arc extraction algorithm based on curve growing is used to extract circular arcs in scale space.Then we put forward the algorithm of dividing the eight quadrant coordinate system to obtain the number of pixels and the coordinates of the pixels of each circular arc.The principal direction of circular arcs and circles is determined.According to whether the content of the image changes,the circular arc support region is generated in the neighborhood of the circular arc using the information of the pixels respectively.We construct circular arc band descriptors in the support region.The descriptor is invariant to scaling,translation,rotation and illumination.At last,matching circular arcs based on the matching criteria by combining of the Nearest Neighbor Distance Ratio(NNDR)and the Euclidean distance constraint of the circular arc band descriptor.We make circular arc matching test experiments in this paper.Circular arc matching results are obtained under different image transformations.The matching results show that the circular arc matching algorithm based on feature descriptor has more correct matches and more than 85% matching accuracy and the running time of the algorithm is short after image blurring,translation rotation transformation,scaling transformation,illumination and viewpoint change,image compression.