Video Caption Detection Algorithm Based on Multiple Instance Learning

摘要

Over the last few decades, multiple-instance learning (MIL) has been successfully utilized to solve the content-based image/video retrieval (CBIR/CBVR) problem, in which a bag corresponds to a video scene and an instance corresponds to a frame caption. However, existing feature representation schemes are not effective enough to use MIL to detect video caption frames from news video, which hinders the practical applications of CBVR. This paper presents an algorithm that regards the video frames containing caption as a bag. It detects, localizes and extracts video caption frames using multiple-instance learning (MIL) automatically. Experimental results show that the proposed method can detect, localize, and extract video caption frames with more high accuracy.