Large-scale video retrieval using image queries
- This dissertation addresses the problem of retrieving videos from large repositories using image queries. Such technology is important for numerous applications, such as brand monitoring or content linking. This work introduces a new retrieval architecture, where the image query can be directly compared to database videos -- significantly improving retrieval scalability, compared to a baseline system that searches the database on a video frame level. Comparing an image to a video is an asymmetric problem, as only videos present a temporal component. Our first contribution is an asymmetric comparison technique for Fisher vector descriptors. We explore problems where query or database items contain varying amounts of clutter, showing the benefits of the proposed technique. Our second and third main contributions consider the design of video descriptors that can be compared directly to image descriptors. We start by constructing Fisher vectors for video segments, by exploring different aggregation techniques. For a database of lecture videos, such methods obtain two orders of magnitude compression gain with respect to a frame-based scheme, with no loss in retrieval accuracy. Then, we consider the design of video descriptors which combine Fisher embedding with hashing techniques, in a flexible framework based on Bloom filters. We investigate different spatio-temporal aggregation configurations, hash functions and scoring approaches. Large-scale experiments using three datasets show that this technique enables faster and more memory-efficient retrieval, compared to a frame-based method, with similar accuracy.
|Type of resource
|electronic; electronic resource; remote
|1 online resource.
|Filgueiras de Araujo, Andre
|Stanford University, Department of Electrical Engineering.
|Guibas, Leonidas J
|Guibas, Leonidas J
|Statement of responsibility
|Andre Filgueiras de Araujo.
|Submitted to the Department of Electrical Engineering.
|Thesis (Ph.D.)--Stanford University, 2016.
- © 2016 by Andre Filgueiras de Araujo
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...