K-means cluster diversified searching method on manifold surface and based on geodesic distance

Abstract

The invention discloses a K-means cluster diversified searching method on the manifold surface and based on geodesic distance. The K-means cluster diversified searching method specifically includes the following steps: extracting features, training and generating a plurality of support vector machines (SVM) with different parameters and choosing an optimal SVM; extracting input image features, and executing searching through the optimal SVM to generate result sequencing; choosing values of a buffering pool and k values adjacent to k by means of a data base (DB) index; performing generic space division on a training set, and adopting an improved K-means cluster method; and obtaining images closest to cluster centers by adopting the geodesic distance, and finally leading out final sequencing. By means of the K-means cluster diversified searching method, the content-based image searching technology is used, recognition and searching on images are automatically achieved, optimal parameters are well chosen, diversification of searching results is achieved under the premise that relevance is ensured, similar or approximately similar searching results are hidden for users, results with representativeness are extracted, and diversified information is supplied to users within time as far as short.

Description

technical field [0001] The invention relates to a diversified image retrieval system, in particular to a geodesic distance-based K-means clustering diversified retrieval method on a manifold surface. Background technique [0002] The research object of image diversification retrieval is: how to make novel, unique, and non-repetitive images rank first in the retrieval results in image retrieval. The application fields of the diversified image retrieval system mainly include: application of image electronic equipment, Internet image search, medical image retrieval, commercial information retrieval, remote sensing information investigation and so on. In recent years, with the rapid development and popularization of digital equipment such as scanners, digital cameras, and digital video cameras, as well as the improvement of multimedia technology and the rapid popularization of the Internet, image data has grown geometrically, so large-capacity Faced with the increasingly huge i...

AI Technical Summary

Problems solved by technology

At the same time, in traditional clustering methods, Euclidean distance is usually used to calculate the distance from a point to the center, which is not suitable for structures with potential manifolds, and there is no suitable index for evaluating the quality of clustering. The present invention will propose solutions to the above problems, and then solve the problem of increasing diversity under the premise of ensuring concept relevance

Images