The Research Of Similar Measurement Method Based On The Hausdorff Distance

Similarity measure is a core technology of the modern similarity science. It is widely applied in the image recognition, the gestures recognition, face recognition and geographical information system (GIS) oriented in areas such as tracking algorithm. By extracting the edge of space target track and transforming the edge into points set of the corresponding space, It is one of the main common method calculating the Hausdorff distance between two space set of points in the similarity measure of modern similarity science.At present, the spatial index technology of common Hausdorff distance computing algorithm adopted the R-trees in order to improve the operation efficiency of algorithm, but the Hausdorff distance accuracy of the algorithm is vulnerable to noise affect. With the growing scale of the data object space, edge track of the space object transformed by space point set also becoming more and huger. Therefore, the performance requirements of the algorithm which calculation of the Hausdorff distance is more and more higher, and how to improve the performance of the algorithm and guarantee the accuracy of the algorithm is an important subject.Based on the above analysis, the paper makes a significant exploration about how to optimize the Hausdorff distance similarity measure algorithm, this paper main works as follows:First, According to the problems which the similarity measure based on Hausdorff distance from the existing in traverse R-trees expenses is larger and the total cost of the execution time slants big, this paper puts forward an improved version of the Hausdorff distance based on the similarity measure. Different from the previous algorithm by R-trees data structure with traverse method for depth first or best first, in this paper, by applied the principle of the branch and bound to the intermediate results are pruned, reduce the times of traverse R-tree number in the algorithm. Simulation results show that this method is significantly, which reduced the algorithm’s time cost, reduce the cost of the traverse for R-trees.Second, The problem of the anti-noise ability in the old algorithm is poor. This paper applies the similarity measure of the Hausdorff increasing distance. It’s different from the basic traverse R-trees and obtain Hausdorff distance, this paper introduce a double direction queue as the intermediate storage structure in the algorithm, increasing access to the R-tree node of different objects. The simulation experiment results show that the improved similarity measure algorithm bases on ncremental Hausdorff distance, ensure the algorithm of the accuracy of similarity measure, and improve the algorithm of the anti-noise ability.Third, INC-HD algorithm is used in the actual civil UAV(Unmanned Aerial Vehicle) remote sensing image processing system, It can get the similarity between two images through calculating the image feature points of the Hausdorff distance value, optimize pretreatment operation of source image data to a certain extent, wash out part of redundancy invalid source image data, enhance the effective image registration, image mosaicing times, so as to improve the system efficiency.