Research And Implementation Of Content-Based Image Collection And Rigional Representative Image Selection

The development of Internet, and the popularization of digital photography, as well as the advent of public media-sharing websites, have led to tremendous growth in community-contributed multimedia resources available on the web. These collections have a previously unimagined depth and breadth, and have generated new opportunities and new challenges to multimedia research. How do we analyze, understand and extract patterns from these new collections and how can we use these unstructured, unrestricted community contributions of media to generate "knowledge" have become a novel research topic nowadays.Aim at the problem above, this dissertation proposed a novel research topic which aims to select the representative photos for regions in the worldwide dimensions. By using a large amount of geotagged images on the photo sharing Web sites such as Flickr, to mine representative photos of the given concept for representative local regions would help us understand how objects or scenes corresponding to the same given concept are visually different depending on local regions over the world.The main content and contribution are listed below. Firstly, we conducted intensive study on Content-based image collection, aimed to build a rich collection of image resources for general concrete concepts. For the image collection, we proposed an effective approach based on content-based clustering on the visual features, by combining global color feature and local bag-of-visual-words feature, to discard most of the irrelevant images and obtain a reduced set of images which are visually similar from each other. Then, Probabilistic Latent Semantic Analysis (PLSA) model is studied, which is recently applied to recognize object categories in an Unsupervised manner. Finally, we select and generate a set of representative images for the representative regions by employing this PLSA model, with the help of geotagged photos collected from the previous step. The results show the ability of our approach to generate region-based representative photographs.In conclusion, the achievement of our research results enriches the approaches on how to make use of these large-scale community contributions of media to generate "knowledge" and improve our understanding of the world, and has a certain theoretical and practical importance. This thesis provides useful methods and approaches for the research and the development of image classification and recognition technology and its applications.