Real-time texture detection using the LU-transform
2006 (English)In: Real-time texture detection using the LU-transform, 2006Conference paper (Refereed)
This paper introduces a fast texture descriptor, the LU-transform. Itis inspired by previous methods, the SVD-transform and Eigen-transform, whichyield measures of image roughness by considering the singular values or eigenvaluesof matrices formed by copying greyvalues from a square patch arounda pixel directly into a matrix of the same size. The SVD and Eigen-transformstherefore capture the degree to which linear dependencies are present in the imagepatch. In this paper we demonstrate that similar information can be recovered byexamining the properties of the LU factorization of the matrix, and in particularthe diagonal part of the U matrix. While the LU-transform yields an output qualitativelysimilar to the those of the SVD and Eigen-transforms, it can be computedabout an order of magnitude faster. It is a much simpler algorithm and well-suitedto implementation on parallel architectures. We capitalise on these properties inan implementation of the algorithm on a Graphics Processor Unit (GPU) whichmakes it even faster than a CPU implementation, and frees the CPU for othercomputations.
Place, publisher, year, edition, pages
Computer and Information Science
IdentifiersURN: urn:nbn:se:kth:diva-61808OAI: oai:DiVA.org:kth-61808DiVA: diva2:479714
ECCV’06 Computation Intensive Methods for Computer Vision Workshop, Graz, Austria, May 2006
QC 201201192012-01-182012-01-182012-01-19Bibliographically approved