The automatic retrieval and analysis of video semantic contents is the fundamental goal and most challenging task in computer vision. And recently, detection and recognition of video objects is one of the most important research areas in video semantic contents analysis. In this paper, we propose a series of related techniques as following: support vector machine, independent component analysis feature extraction, support vector machine and independent component analysis hybrid learning scheme and so on. We also classify human faces in videos and images according their semantic importance in real application, give the definition of 'semantic face' and detection algorithm, and apply 'semantic face' in video news indexing.The paper will be organized as below: In chapter 1, we firstly define several important concepts related with video semantic content analysis and survey the main progress in this area. We also give the definition of video objects discussed in this paper. In chapter 2, we introduce the video/image caption detection techniques. After survey the traditional algorithms, we propose that besides the searching of best video features of caption, the generalization of classifier is also a very important aspect. So we use support vector machine as the classifier, combined with several other steps such as feature extraction, pyramid model and post-process to realize the video caption detection with limited samples. In chapter 3, we focus on the feature extraction problem in video object detection. We present a hybrid learning scheme that firstly using independent component analysis technique to extract the features of video object image sub-block and then using support vector machine to make decision based on these features. We then apply this scheme in detection of human face in video/image. In chapter 4, we discuss the different semantic importance among different human faces and propose the definition and detection flow of 'semantic face'. Then we realize anchorperson shot detection and video news indexing based on semantic faces, which shows video object's important usage in analysis of video semantic contents. In chapter 5, we introduce the process of our algorithms, and present lots of experiment results and discusses. In chapter 6, we present the breakthrough of this paper, some problems still need to solve in our research work and forecast the future of video based human animation techniques. |