Beta 1


Title Object Recognition
Author Bolvig Stentebjerg, Peter Rene
Supervisor Aanæs, Henrik (Image Analysis and Computer Graphics, Department of Informatics and Mathematical Modeling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
Institution Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
Thesis level Master's thesis
Year 2008
Abstract This thesis explores a method to do object recognition on the basis of the visual representation of the bag-of-words model. This approach has shown promising results on large-scale image retrieval systems. The bag-of-words model arises from Internet-based document search and is hereby based on the visual analogy to such application. The method is defined by first extracting local image features using the approach of SIFT, and afterwards hierarchically quantise those features into a vocabulary tree. In this way the images are categorized into representatives of so-called visual words, which give the opportunity for an efficient image search on a large-scale image database. The technique is further extended by retrieving essential color information from the images. This color information is appended to the SIFT-descriptor, after which PCA can be applied to reduce the dimensionality of the joined descriptor. This has shown a successful improvement of the results. Furthermore an estimation on the geometric relation between the query image and the images retrieved are examined on the basis of epipolar geometry. The evaluation on this geometry is then used as a constraint to rerank the list of retrieved images. The image database created for this project contains several images of buildings and statues representing a wide range of tourist attractions in Copenhagen. The objective is to evaluate low quality images, taken with the built-in camera of a mobile phone belonging to the common tourist, and retrieve images from the database based on the recognition to the specific tourist attraction. A sample of test images taken with such a low quality mobile phone camera have been examined on the image database, which produce satisfying results on the recognition of all tourist attraction.
Series IMM-M.Sc.-2008-73
Fulltext
Original PDF ep08_73_net.pdf (8.51 MB)
Admin Creation date: 2008-08-15    Update date: 2008-08-15    Source: dtu    ID: 222467    Original MXD