Semantic Concepts Extraction Research Based On S_T-Simfusion Algorithm And Ontology

Video data contains a wealth of multimedia information,including images, audio and text and so on. The information is concerned with people's daily life, work, leisure and entertainment. So the retrieval research on video data is becoming more important. However, because of the soaring in number and the complexity in contents and structure of the video data, the use and management of video data have become more difficult. In order to query, organize and use the video data effectively, in recent years, video retrieval techniques, especially that retrieval techniques based on semantic concepts have been given to attention by researchers and is a hot in the field of study about multimedia data retrieval.Based on comprehensive analysis some existing algorithms for video semantic concepts extraction, the paper presents the method which combines with ST-Simfusion algorithm and ontology. After above researches, we implement a prototype system. The main contents are described as bellows:(1) Aiming to the shortages (e.g. subjective,Irregular, weak Generality) for labeling samples in the presenting ordinary video semantic retrieval methods, the article proposes a method based on ontology concepts for labeling sample to improve the generality of semantic extraction model;The paper also proposes a new text measure based on ontology concepts, namely, using the ontology concepts similarity as texts similarity to improve the execution efficient of algorithm.(2) A novel approach which combines with the clustering algorithm based on adaptive thresholds ST and Simfusion is proposed to account the similarity among shots, namely, ST-Simfusion algorithm. The method extracts the key frames using the clustering algorithm and account the shots similarity according to the Simfusion algorithm idea.The method is beneficial to the accuracy of semantic concepts extraction, because it not only makes sure information integrality of shots, but also makes full use of associated co-occurrence characteristic among multi-modal in calculating shots similarity.(3) The paper proposes a new algorithm based on ST-Simfusion algorithm and ontology to train semantic extraction model. Firstly, the ST-Simfusion algorithm is applied to shots clustering for getting the shots similarity matrix. Then, using LPP algorithm to lower the dimensions of the shots for getting corresponding coordinates; At last, train semantic concepts extraction model with the coordinates and the shots similarity matrix as input parameters.(4)The paper implements the prototype system of semantic extraction model adopting the 00 principles. The system is composed of function modules, including video data pre-process modular, features extraction modular, shots similarity calculation modular, semantic concepts extraction modular and so on. Experimental results are given for verifying the effectiveness of the method presented in this paper.