Japanese / English

Detail of Publication

Text Language Japanese
Authors Koichi Kise, Masakazu Iwamura, Tomohiro Nakai, Kazuto Noguchi
Title Large-Scale Image Retrieval Based on Hashing of Local Features
Journal Proc. First Forum on Data Engineering and Information Management
Presentation number D6-3
Reviewed or not Not reviewed
Month & Year March 2009
Abstract Image retrieval with query images is a task to find their corresponding images (near duplicates) from the database. It is an important task since it has many applications such as a replacement of barcodes and copyright protection of images. The task of retrieval is not trivial since query images are often far different from their corresponding images in the database. This is due to various imaging conditions, e.g., occlusion, perspective distortion, different lighting conditions, deforcus and motion blur. In order to solve this problem, retrieval based on local featuers attracts considerable attention. In this method, images are indexed using feature vectors extracted from local regions of images. Although they allow us the retrieval relatively robust to imaging conditions, they pose a problem of processing time since the number of local features is huge. In this report, we describe our new retrieval methods based on hashing for solving the above problem. As images in the database, we deal with ordinary images about scnene, objects and people, etc., as well as document images that are completely dissimilar to ordinary images. Local featuers for each type of images and the common notion for successful retrieval are shown.
Back to list