Content-based Multimedia Browser using Locally Linear Embedding

Abstract

Many people today are using digital camera to take photos and store their photos in the computer. Due to increasing numbers of photos, people may hard to find a particular photo out of the photos pools. The conventional photo browser only provide functions in classifying their size, file name and modified day ect. What if a user needs to separate the photos that were taken at day time and night? Why users cannot classify the photos in their content similarities. Obviously, the traditional photo's browser does not provide support for classifying photos by their content. To cope with existing needs, a Content-based Multimedia Browser using Locally Linear Embedding (LLE) is created in this project. LLE is an unsupervised learning algorithm that computes low dimensional, neighbourhood preserving embeddings of high dimensional data. After applying the algorithm to a multimedia collection, users can explore the collection graphically with the Multimedia Browser, where items with similar content are placed next to each other. In the browser, Classifying function will be enhanced to support grouping the multimedia by its content and allow to search by simple keywords.