Content Based Image Retrieval And Video Annotation

With the rapid advances in computer science and technology,new technologies like communication network,data compression and high-capacity storage provide enormous contact with multimedia information to people.Readable multimedia information is stored on various media and forms a large information pool.Different innovations are also influencing the way people get multimedia information.For example,people can get pictures and videos by digital cameras and digital camcorders,and it requires extraction of information structure and semantic features to annotate,model and organize these information.To get useful information from the large amount of information,multimedia information retrieval has become an important research topic.Our research focuses on content-based image retrieval and video annotation.There are various sources for images including downloaded pictures from the internet and self-shot images from cameras.Early image retrieval technology usually uses the manually-annotated information for images as index to retrieve related images in the database,which is not applicable for huge amount of data,not only for the lack of human power to do so,but also due to different understanding of the image content by different people.Now content-based image retrieval is a popular research area and many practical image retrieval systems have been developed.This paper researches about query based image retrieval and has proposed two new methods for similarity measurement.Referring to another important content of multimedia information, video,there are more and more requirements for video retrieval these days due to rapid development of online video resources.Video can be considered as the combination of image,audio,text and animation. For video retrieval,annotation is an important step.Also,manual annotation is an impossible mission due to the huge manual labor and different understandings to the same image,so people are working on to find automatic annotation methods for videos.Another challenge is to save effort when annotating videos.Methods based on compressed MPEG videos are good choices.Our research is also focused on the way to develop a semi-automatic annotation method for compressed MPEG videos.Image retrieval and video annotation are both based on the extraction of features for multimedia sources.During the past years, people have proposed many features including color,texture and shape,etc.For one type of features there are also numerousrepresentations.With the advances of research,MPEG-7 standards have been proposed.These standards were developed since 1996 and they provide description scheme for multimedia contents.The visible description tools provide standardized description for image contents and can enhance the generality of image feature description.In 2001,MPEG-7 standards were officially released and became the international standards.Our research has applied several MPEG-7 features to increase the accuracy of retrieval.Content based multimedia retrieval technology is becoming more mature and it possesses influential social value.Combined with traditional database technology,it can realize the storage and management of huge amount of multimedia data;combined with web searching engine technology,it can be used to retrieve useful information from web pages;combined with digital camera,digital camcorders,it makes it possible for people to manage their own multimedia contents;combined with IPTV,it makes video and image retrieval accessible through these new user interfaces.In the foreseeable future,content based multimedia retrieval technology will be more widely applied.This paper demonstrates our research work with the topic of content-based image retrieval and video annotation,introduces image segmentation based feature extraction,similarity measurement algorithms,compressed MPEG video's camera motion vector extraction and video annotation technologies.The first chapter introduces developments in the field of multimedia retrieval and introduces several representative image retrieval systems and video annotation systems.Then we list several key technologies and main developments in this field.The second chapter introduces details of important technologies of content-based image retrieval including image feature extraction,feature's similarity measurements.In the process of image feature extraction,besides features defined in MPEG-7,we implement image segmentation using Mean Shift algorithm and thus provide more information about segmented regions of the image for further processing.The third chapter introduces our innovative work of content-based image retrieval,mainly focusing on the improvement of similarity measurement.We have proposed two new methods for feature similarity measurement.In the fourth chapter, we present our research work in the field of video annotation.This chapter gives the introduction of MPEG compressed video formats and compare the extraction of camera motion vector from compressed MPEG videos and uncompressed MPEG videos.In the fifth chapter, combined with video structure analysis,we propose a semi-automatic sports highlight annotation method and give detailed demonstration.In the last chapter,chapter six,we make a conclusion and give some discussions about future work.