Video library index table establishing method and device, server and storage medium

Abstract

The invention discloses a video library index table establishing method and device, a server and a storage medium. The method comprises the steps of performing feature extraction on a target video frame of each video in the video library to obtain a feature vector of the target video frame; dividing the feature vector into a plurality of feature sub-vectors corresponding to the plurality of firstsegments according to a preset first dimension division strategy; clustering the feature sub-vectors corresponding to each first segment, and determining a segment center vector of each first segment;determining a Cartesian product between the segmented center vectors to obtain a library center vector set; for any target video frame, determining a target library center vector according to the distance between the target video frame and the library center vector; determining a quantization vector according to the target library center vector; and determining a first index field and a second index field corresponding to the target video frame in the index table according to the target library center vector and the quantization vector. According to the method, the training time consumption and the iteration period of the codebook vector are shortened, and the construction efficiency of the index table is improved.

Description

technical field [0001] The invention relates to the field of computer technology, in particular to a method, device, server and storage medium for establishing an index table of a video library. Background technique [0002] With the rapid development of video services, the search efficiency requirements for related videos are getting higher and higher. When searching for related videos, it is generally based on the index table of the video library. When the index table is established, the video fingerprint (the feature vector of the video frame) of the video in the video library is first obtained, and then the full-dimensional video is analyzed by a clustering algorithm. The fingerprint is trained to obtain the codebook vector of the video library, and then the index table is constructed through the codebook vector. [0003] It can be understood that the longer the codebook vector of the video library is, the less video data there is in each bucket (that is, the cluster co...

AI Technical Summary

Problems solved by technology

In the related art, the index table is constructed based on full-dimensional video fingerprints. When a longer codebook vector is to be obtained, the amount of calculation of the system will be greatly increased, and the training time and iteration cycle of the codebook vector will be greatly extended, which will lead to index Table construction is inefficient

Images