P. Perumal , D. Anandhu
In this paper the video retrieval process is evaluated to produce the top ranked search results to the query relevance. i.e., every large search engine log shows that the users are really interested in top ranked result according to their query. Therefore, it is essential to achieve high accuracy in video search retrieval. Generally, the search query for video retrieval is converted to text query and then searches for the relevant results (videos). While many methods exist for improving video search performance, they pay less attention to the above factor or encounter difficulties in practical applications. To overcome the limitations of the existing reranking methods we present a flexible and effective method called Cross Reference Reranking (CR- Reranking), to improve the retrieval effectiveness. To provide high accuracy in video retrieval, CR-Reranking involves a cross reference method, to fuse multimodal features. Particularly, multimodal features are first taken separately to rerank the initial search results at the cluster level with cluster number, and then all the ranked clusters from different modalities are fused together and produce the top ranked search results with high relevance to user query.
Clustering, Fusion, Pseudo relevance feedback, Reranking, Relevance feedback.