TUTCRIS - Tampereen teknillinen yliopisto

TUTCRIS

Model structure selection for sparse predictive coding utilizing warped views

Tutkimustuotos

Yksityiskohdat

AlkuperäiskieliEnglanti
OtsikkoProceedings of the Eighth Workshop on Information Theoretic Methods in Science and Engineering
ToimittajatJorma Rissanen, Peter Harremoes, Soren Forchhammer, Teemu Roos, Petri Myllymäki
JulkaisupaikkaHelsinki
KustantajaUniversity of Helsinki, Department of Computer Science
Sivut7-10
Sivumäärä4
VuosikertaReport B-2015-1
TilaJulkaistu - 1 joulukuuta 2015
OKM-julkaisutyyppiD3 Artikkeli ammatillisessa konferenssijulkaisussa
TapahtumaWorkshop on Information Theoretic Methods in Science and Engineering -
Kesto: 1 tammikuuta 1900 → …

Conference

ConferenceWorkshop on Information Theoretic Methods in Science and Engineering
Ajanjakso1/01/00 → …

Tiivistelmä

The color image encoding of one view, conditional on the other view, makes use of the disparity image for comput- ing a warped partial approximation, to be used for sparse predictive coding. Different estimates of the disparity im-
age, produced by different stereo matching algorithms, will result in different final codelength values for the conditional coding of the color image. In here we study how the ranking of stereo matching algorithms based on the predictive coding codelength, which does not need the knowledge of the ground-truth disparity, relates to the other rankings of the stereo matching algorithms, that require the availability of the ground truth disparity. We show that the two information-theoretic variables introduced herein correlate well with the most used quantity for stereo matching ranking.

Julkaisufoorumi-taso