Professional paper

Application of SVD decomposition of a matrix in information retrieval and image compression

Zvonimir Ivančević ; Privredna banka Zagreb
Slobodan Jelić ; Odjel za matematiku, Sveučilište J.J. Strossmayera u Osijeku, Osijek

Abstract

This paper gives the application of singular value decomposition (SVD) in low-rank approximation of a matrix. Image compression is performed by taking only first k singular values in SVD of the matrix.
Illustration of information retrieval is based on the concept of a latent semantic indexing model (LSI). The database is determined by the matrix whose columns represent documents. The database matrix
is substituted with a low rank-approximation matrix, while searching for relevant documents with respect to the user’s query is based on calculating the cosine of the angle between query and document vectors. The smaller the angle, the more relevant is the document for the given query.

Keywords

singular value decomposition; image compression; information retrieval; vector space model; latent semantic indexing; RGB image; monochrome image; rank of the matrix

Hrčak ID:

165818

URI

https://hrcak.srce.hr/165818

Publication date:

1.8.2016.

Article data in other languages: croatian

Visits: 1.852 *