Video Semantic Detection Method Based On Gaussian Mixture Model Visual Feature

Recent years, the amount of consumer-generated videos we can access over the Internet has been rapidly increasing since the development of Internet and multimedia technology. Video has become a major source of content on the WWW, Digital TV and other multimedia application fields. The rapid increase in the available amount of video data is creating a growing demand for efficient methods for understanding and managing it at the semantic level. Semantic is the meaning of data, video semantic analysis is essential in video indexing and structuring. In particular, detecting semantic depicted in a video enables us to get significant information. At present, how to help users to retrieve information they need quickly and accurately in the huge amount of video data has become the urgent problem to be solved. To overcome the semantic gap between low-level features and high-level semantic concepts, video semantic modeling has become a research focus in the field of video management,classification and retrieval. Therefore, video semantic analysis and research are very important.Firstly, the background and significance of the subject are introduced in the thesis,the research status of video semantic analysis and the existing problems are also discussed. Secondly, some key technologies in video semantic detection such as video segmentation, key frame extraction and feature extraction are briefly described. Finally,based on the current theoretical research on video semantic analysis, this thesis presents a video semantic detection method based on topographic independent component analysis and Gaussian mixture model and then puts forward the research on Gaussian cloud mixture model for video semantic detection. A video semantic analysis prototype system is also developed. The main contents are as follow:(1)The video semantic detection method based on topographic independent component analysis and Gaussian mixture model is proposed. Firstly, features of each video clip are extracted by topographic independent component analysis algorithm, this method can learn complex invariant features from video clips. Secondly, the featuredistribution of each video clip is described by Gaussian mixture model. Finally, a Gaussian mixture model supervector is created for video semantic detection. The experiment results show that this method can overcome the human factors from current manual feature extraction algorithm and reducing the quantization error in the process of modeling. In particular, this method effectively improves the accuracy of video semantic detection.(2)The video semantic detection method based on Gaussian cloud mixture model is put forward. The process of traditional Gaussian mixture model does not consider the uncertainties of concept such as randomness and fuzziness. Cloud model is able to deal with the randomness and fuzziness of video concept itself and to describe the semantics of video in a manner of human understanding. Gaussian cloud is one of the cloud models and has universal adaptability. Using cloud distribution in the mixture model instead of Gaussian distribution and combining Gaussian mixture model with Gaussian cloud model provide an idea to realize the transformation from quantitative data to qualitative concept and gives a good solution to solve the ambiguity of video concept. Experimental results show that the application of this method to video semantic detection can improve the noise tolerance capability so that the model has better generalization ability and the accuracy of video semantic detection is also more desirable.(3)Using the object-oriented design method and the mixed programming model of MATLAB and C#, we design and implement a video semantic detection system based on Gaussian mixture model visual features to verify the effectiveness and availability of the algorithm.