Research On Geolocation For Images Based On Hadoop

With the development of multimedia technology and Internet, photo-sharing websites are increasingly popular. Due to the participation of users, photo-sharing websites have stored huge amounts of multimedia informations, which contain a lot of images, text tags and user informations. Some images contain GPS coordinates and text tags also contain geographical location description informations. User-related GPS coordinates are of great value for research and location-based search are increasingly popular too, so there is a broad application foreground for geo-tagged images.Massive geo-tagged images can provide valueable informations, but the percentage of images with accurate geographical location among online images is too low. Whereas manually annotated geographical locations are often inaccurate, so it is meaningful to estimate other images’ geographical location using geo-tagged images. Meanwhile, the requirements of processing massive images bring higher requests for applications’ abilities of data storage and processing. Traditional data processing technologies have been not adapted to the requirements of processing massive images. And cloud computing provides a new approach to store and process massive images.This dissertation analyses the research background and current status at domestic and abroad, and studies the image file storage scheme based on Hadoop. With understanding and analysis of the problems of processing small files based on Hadoop as well as analysis of the existing solutions, this dissertation proposed an improved image file storage scheme and designed the storage access interface. The improved scheme optimizes the storage of small files by merging files. This dissertation also analyses the tag-based and content-based geolocation solutions for images, as well as studies the key technologies such as:clustering of GPS coordinates, classification of text tags, image features extraction and similarity calculation. Based on above analysis and studies, this dissertation proposed an improved geolocation scheme for images based on Hadoop, and implemented an image geolocation system based on Hadoop using Java programming language and SQL Server2012database as well as the baidu map API.The improved scheme supports merging operation and appending operation of image files, which facilitates the management and processing of image files. The improved scheme distributes images into some regions by using GPS coordinates clustering technology and text tags classification technology, and combines the text similarity with image similarity, which can take effective use of image and tag informations. The final experimental results show that the improved image file storage scheme has a good storage access performance and the improved geolocation scheme for images has a relatively higher accuracy than other schemes. The system running results also achieved the expected effect.