| Due to rapid exponential growth in data, a couple of challenges we face today are how to handle big data and analyze large data sets. An IBM study showed the amount of data created in the last two years alone is 90% of the data in the world today. We have especially seen the exponential growth of images on the Web, e.g., more than 6 billion in Flickr, 1.5 billion in Google image engine, and more than 1 billon images in Instagram. Since big data are not only a matter of a size, but are also heterogeneous types and sources of data, image searching with big data may not be scalable in practical settings. We envision Cloud computing as a new way to transform the big data challenge into a great opportunity.;In this thesis, we intend to perform an efficient and accurate classification of a large collection of images using Cloud computing, which in turn supports semantic image searching. A novel approach with enhanced accuracy has been proposed to utilize semantic technology to classify images by analyzing both metadata and image data types. A two-level classification model was designed (i) semantic classification was performed on a metadata of images using TF-IDF, and (ii) image classification was performed using a hybrid image processing model combined with Euclidean distance and SURF FLANN measurements.;A Cloud-based Semantic Image Search Engine (CSISE) is also developed to search an image using the proposed semantic model with the dynamic image repository by connecting online image search engines that include Google Image Search, Flickr, and Picasa. A series of experiments have been performed in a large-scale Hadoop environment using IBM's cloud on over half a million logo images of 76 types. The experimental results show that the performance of the CSISE engine (based on the proposed method) is comparable to the popular online image search engines as well as accurate with a higher rate (average precision of 71%) than existing approaches. |
