| Abstract [eng] |
In this paper the problem of image search is addressed. The primary goal of the paper is to present a prototype that can be used to identify and effectively return search results to the end user (i.e., return the most relative images to some given search criteria). All the search is performed to find images stored in one personal computer. To accomplish this task an image rank algorithm is utilized. Algorithm's advantages and disadvantages are pointed out. As a result an optimization issue is discussed and a way to increase the performance of multidimensional data indexing is presented. The optimization is based on KD trees. The assumption is that these trees might work faster than ball trees. Those were proposed by authors of the original image rank algorithm. The result is a prototype that is used for searching and presenting images based on ranks that indicate relative importance of each of the image to some given search criteria. The results are prioritized. The prioritization might be thought of as sorting images by their similarities. By using the application a practical analysis was performed and the results were described in details. It was found that KD trees can be effectively used to index multidimensional data while calculating image ranks. It not only simplifies the algorithm but also provides better performance results than ball trees. |
