Study On Image Matching Algorithm Based On Image Contour And Its Applications

Image matching is one of the important issues in computer vision and remote sensing image fusion and navigation, that is, to align the different images of the same scene. Under the assumption of similarity transformation between two curves, this paper studies the image matching and stitching. The main contents are summarized as follows:(1) Firstly, A brief introduction to the meaning and purpose of the image matching technique is showed, and the mathematical definition of the image matching and the 2D geometry transformation model are described. Secondly, some matching algorithms are reviewed which are respectively intensity-based ones, Fourier-Mellin transform-based ones and feature-based ones. With these algorithms, feature-based methods are widely used. One of the advantages of these approaches is to improve the computational efficiency. However, evidently, the accuracy of the feature extraction will affect the match result directly.(2) In the paper, a new curve matching algorithm based on the V-system is proposed which avoids the difficult to extract feature points. V-system is a class of new non-continuous complete orthogonal function system in L2 [0,1]. It would be accurate to say that V-system is a class of piecewise polynomial formed by the orthogonal functions which includes not only discontinuous functions but also continuous ones. Recently, some good results based on V-system have been obtained in orthogonal geometry expression, pattern recognition, point cloud data, digital watermarking and so on. Here, based on the assumption that the two curves satisfy the RST transformation, that is, similar transformation (R:rotation, S:scale, T:translation), the matching algorithm based on V systems is proposed. Contour features are extracted firstly from one image to be considered as a whole feature curve, and it is no need to extract further local feature of the curve. Through lots of simulation and real tests, the result shows that the algorithm is feasible.(3) Based on the above algorithm, the two images are stitched together which can be taken by an ordinary digital camera. In spite of that the images do not satisfy the RST transformation strictly, the stitching result is satisfactory which shows the feasibility of our algorithm again.