Indexing images of buildings based on geometrical invariant Hough descriptors
Li, Chang-Tsun and Yuan, Xiang (2012) Indexing images of buildings based on geometrical invariant Hough descriptors. Coventry: University of Warwick..Full text not available from this repository.
Official URL: http://eprints.dcs.warwick.ac.uk/id/eprint/1579
A CBIR system for retrieving images with specific buildings from databases is proposed in this paper. We exploit the parallelism invariance property of the line features of buildings in images in the derivation of two new geometrically invariant linear feature descriptors. We call these descriptors Hough descriptors as they are extracted from the Hough transform domain. The underlying concept is to utilise the invariance of parallel line features such that the individual edges (local property) can be used in a collective manner to embed their global relationship in the feature descriptors. Upon receiving a query image, the CBIR system transforms the edges detected from the query image into the Hough transform domain. The transform domain is divided into 180 degrees/bins in order to reveal the linear edge distribution. From each bin, the peak percentage profile and distance ratio profile are calculated to serve as the descriptors of images. That is to say that an image descriptor consists of two components (peak percentage profile and distance ratio profile), each with 180 elements. The circular correlations between the peak percentage profile and distance ratio profile of the query image and those of the database images are then taken as the similarity measure for ranking the relevance of the database images to the query.
|Subjects:||Q Science > QA Mathematics > QA76 Electronic computers. Computer science. Computer software
?? QA76.73 ??
|Divisions:||Faculty of Science > Computer Science|
|Publisher:||University of Warwick|
|Place of Publication:||Coventry|
|Number of Pages:||24|
|Institution:||University of Warwick|
|Theses Department:||Department of Computer Science|
|Status:||Not Peer Reviewed|
Actions (login required)