Image Retrieval, opening up the bag-of-words
This course gives the theory and practical exercises needed to build a state-of-the-art bag-of-words image-retrieval system. Specifically, the course focuses on the methods necessary for particular-instance retrieval, which is the task of ranking the most visually similar instances of a query object (e.g., art work, landmark, or logo) from a very large-scale image database, typically tens-of-millions of images or more.
K-nearest-neighbor classifiers and approximate nearest neighbor algorithms, vector quantization, min-hash, query expansion, spatial verification and the design of local features for image representation. Several fundamental concepts from computer vision are integral to the topic list and will be covered as well.
The main programming tool for the course is Matlab. The organizers will provide the students with a virtual machine with preinstalled software. Script templates will be provided by the lecturer.
The course is self-contained, but the student is expected to have a basic familiarity with statistics, machine learning, and data structures. Matlab will be used for the practical exercises.
James Pritts is a researcher at the Center for Machine Perception (CMP) and a PhD candidate at Czech Technical University. His current research is detecting and modeling repeated patterns in images. Prior to joining CMP, Mr. Pritts had an industry career in computer vision and scientific computing. Mr. Pritts was a Lead Engineer for BAE Systems, where he contributed to several US Department of Defense (DARPA) computer-vision research efforts; at NASA, he created gesture-recognition software for remotely controlling robotic arms of the International Space Station; and for Shell Global Solutions, he designed high-performance process-control algorithms. He received his BSc in Mathematics at The University of North Texas and his MSc in Computer Science from Czech Technical University.