Capitalizing The Collective Knowledge for Video Annotation Refinement using Dynamic Weighted Voting

Abstract

Collective knowledge paves the way for most challenging task to be an interesting and improving efficiency in the field of multimedia annotations and retrievals. Automatic annotation is bridging the gap between low-level content and high-level semantic concepts. It has been an active research area in the field of multimedia retrieval, machine learning and social media environments. Even most automatic annotation approaches are often unsatisfactory, the annotation refinement has invited the attention of recent researchers. In this paper, a novel refinement algorithm is proposed using dynamic weighted voting based on mutual information. It leverages the collective knowledge of the social media like collection of videos, images, texts in the form of tags, and comments available online. The proposed algorithm invests collective knowledge to measure the relevance between the candidate annotations by assessing the probability and assigning a dynamic weights.