User friendly approach for video search technique using text and image as query

摘要

With the evolution of Internet and Computer Technology, the digital videos are increasing explosively. There's an enormous number of videos available online. So how to get the interesting video clips from the massive video dataset quickly and efficiently has become an urgent problem in the field of content-based information retrieval. Content-Based searching and retrieval of video data has become a challenging and important issue among the search engines. Generally videos are retrieved from a large collection of videos based on the keyword which is only text. "Multimodal Fusion for Video Search Re-ranking" is a flexible and effective reranking method, and it is also called CR-Reranking, to improve the retrieval effectiveness. To offer high accuracy on the top-ranked results, CRRe-ranking employs a cross-reference (CR) strategy to fuse multimodal cues. Given a text query by users, the system then returns a series of approximately relevant video shots of matching the input text with the text documents associated with the video shots. But the results of this method are less relevant as the video contains only text in the title. Also this method has time overhead. In the proposed system, videos are retrieved not only through keywords but also using an image as a query. Users are usually interested in the top ranked portion of the returned search results and therefore it is crucial for search engines to achieve accuracy on the search results. Content based video searching and retrieval framework is proposed to improve the efficiency and accuracy of the search engines. The input is given as a video at the training time, where the video is converted into images. At the time of searching, user gives a text or an image as a query to the system which then returns a series of additional relevant video shots than the previous methods by mapping with the images. At the time of training, the user can associate some relevant text to the extracted images and then some query image- can be used to search the relevant video content. After implementing the proposed system the performance analysis of the system along with the memory and resource consumption has also been performed. After comparison it was found that the proposed system provides much better performance than other previously designed systems.