Research on optimized R-Tree high-dimensional indexing method based on video features
Ningning Chen, Xian Zhong, Lin Ji Li · 2017
In order to solve the dimension disaster problem of Video high dimensional feature, a new indexing method is proposed: PKSR-Tree index. PKSR-Tree index first uses the principal component analysis to reduce the dimensionality of the high-dimensional feature data, reducing the dimension of the disaster impact and making the distribution of data homogeneous. The feature data after dimensionality reduction are divided by k-means clustering. It reduces the error of K-means algorithm and solves the shortcomings of the K-means algorithm which is susceptible to noise and unable to find data other than spherical shape. The SR-Tree index is established for each cluster, which reduces the problem of multi-path query. The searching process makes use of clustering partition to reduce the scope of candidate results, so that the searching range can be quickly filtered. Experiments show that PKSR-Tree index has obvious advantages in the efficiency and accuracy of video high-dimensional feature data.