Machine Learning for Video Classification and Retrieval

Alexander Hauptmann Dept. of Computer Science Carnegie Mellon University Pittsburgh, PA 15213 USA Video analysis and retrieval from video collection is a difficult task. One would like to characterize the video in terms of various types and styles, understand what objects are in the video, divide it into camera shots and group those into coherent scenes. Ultimately we want to make video as tractable and ‘searchable’ as current text collections that are indexed through the web. Video analysis is orders of magnitude more complex than speech recognition, where the data stream is merely a one-dimensional signal and suitable intermediate level representations such as sentences, words, and phonemes permit a divide-and-conquer approach utilizing machine learning.