Efficient Bulk-Insertion for Content-Based Video Indexing

Abstract

Videos have become one of the most important communication means these days. In this paper, we propose an approach to efficiently bulk-insert a set of new video index-entries into the existing video database for content-based video search. Given the current situation that enormous amount of new videos are created and uploaded to the video sharing websites, the efficient approaches are highly required. The environment we focused is where a B⁺-tree is applied to index the video content-features. We propose a hybrid bulk-insertion approach based on a well-known bulk-insertion. Unlike the traditional bulk-insertion in which the traversals to insert the remaining index entries are performed to the ancestors, we propose to add a leaf-level traversal to improve the efficiency. Thus, our approach works in a hybrid manner, i.e., it switches between the leaf and ancestor traversals with regard to a condition with a very small additional cost. The experiments have been conducted to evaluate our proposed work by comparing to the one-by-one insertion approach, and the traditional bulk-insertion approach. The experiment results show that the proposed approach is highly efficient for video content-based indexing.