Research Of Multidimensional Index For Medical Image

The application of Multimedia especially image resource in life and work is more and more wide, due to the rapid development of Internet and information technology. And with the advance of medical technology, a great deal of medical imaging is produced. So content based image retrieval system is used in medical image retrieval. Now,how to search image data efficiently and accurately become a problem.A common method is to extract image feature, and then compare their characteristic. A comprehensive description of multiple variables is more easy to reflect the characteristics of the original image resources. So the extracted feature is multidimensional data. Obviously with the dimensions increase, the mapping from image to feature data become more reasonable and accurate. The traditional data structure and retrieval algorithm is designed for two-dimensional space or three-dimensional space, and can’t meet needs because medical image data is multidimensional.This paper studies the characteristics of medical image, and put forward the specialized index. First proposing index based on kd-tree. Bisecting data approximately to solve the problem that tree is imbalance. Proposing a new method to select divided dimension, so data can be split along the best axis each time. According to the characteristics of data, a new algorithm used to adjust tree is proposed to alleviate imbalance of the tree caused by insertion. Then proposing improved index based on R-tree. On the basis of image characteristics, raise a new division algorithm and improve construction algorithm, so index can be constructed without coincidence. Using the Heuristic best-first traversal instead of using breadth or depth traversal to avoid loss of efficiency caused by search order. In addition when dimension is too high, original dimension are replaced by a set of suitable dimension which is selected by a large number of experiments. Pruning property ensure the correctness of query. At last, identify the scope of each index by conducting experiments and verify the correctness and efficiency of improved index.