Research On Movie Similarity Calculation Using Multi-feature Method

Along with the rapid development of computer networks and multimedia technology, video and all kinds of multimedia data is growing exponentially, how to search similar videos from huge video resources has become a common concern of Internet users. Movies as video resources are the most common form of media Internet users familiar with, and gradually become one of the largest groups of Internet video. Therefore, the research of large-scale movie similarity calculation method has been a very important significance for solving video similarity in the entire Internet.The existing video similarity calculation method is based on the multi-video low-level features extraction and processing technology, greatly differs from the high-level semantic concepts that people can understand, this has significantly affected the calculation of video similar in the actual result. How to obtain the video semantic information of these films to rid of the underlying characteristics of the video will become an important research method. So we can not only focus on the extraction of local text’s expressive feature, but also on the overall information of the movie. This can help us calculate the similarity of the movies better and in a more objective way.In this dissertation, movie similarity is our research object and the video information in the Internet Movie Database (IMDB) is the source of research data. Then it analysis the strengths and weaknesses of existing similarity calculation method for videos through learning and research based on the lower features and video-based key-frame matching of video similarity research methods. After that it discusses the effects of keywords and text descriptions in the calculation of video high-level semantic similarity and the feasibility of the method used the combined information. Moreover, in the movies based on feature classification, the law of distribution of these characteristics, the improvement of word frequency in a text information extraction process-inverted document frequency statistic method and experimental verification of the validity of the algorithm, and completed the following research work: 1) In order to analyze some of the existing video of the underlying characteristics and key-frame matching of video similarity method to calculate the advantages and disadvantages, as well as the applicability of these methods in the field of film similarity, and then the analysis based on the movie of the keywords and description text Film similarity algorithm, and its feasibility is verified by experiments.2) According to the classification characteristics of the properties of the film keyword optimized text similarity computation feature selection strategy, improve the similarity of the film to calculate the accuracy of the text feature extraction process.3) Movie similarity calculation, based on the dynamic changes of text vector in the video high-level semantic similarity with the keyword matching, adjusting the coordination parameters, to ensure the correct rate of reduction rate and results of the optimal algorithm.