TUTCRIS - Tampereen teknillinen yliopisto

TUTCRIS

Comparison of Cost Aggregation Techniques for Free-Viewpoint Image Interpolation Based on Plane Sweeping

Tutkimustuotos

Yksityiskohdat

AlkuperäiskieliEnglanti
OtsikkoNinth International Workshop on Video Processing and Quality Metrics for Consumer Electronics
AlaotsikkoVPQM 2015
TilaJulkaistu - helmikuuta 2015
OKM-julkaisutyyppiD3 Artikkeli ammatillisessa konferenssijulkaisussa
TapahtumaInternational Workshop on Video Processing and Quality Metrics for Consumer Electronics -
Kesto: 1 tammikuuta 2000 → …

Conference

ConferenceInternational Workshop on Video Processing and Quality Metrics for Consumer Electronics
Ajanjakso1/01/00 → …

Tiivistelmä

Free view interpolation based on plane sweeping is a novel image-based rendering technique which allows rendering virtual view at arbitrary position, directly from a set of calibrated and non-rectified input images. The method does not require pre-estimated depth maps or other geometric information, and hence is attractive for real-time applications. Similarly to classical stereo matching, it relies on finding color and intensity similarities between pixels at assumed depth hypotheses and therefore is prone to inconsistencies in the calculated dis-similarity values. These are treated like noise in the cost volume domain and subsequently mitigated by proper cost volume filtering. The operation is also known as cost volume aggregation. In this work we compare the performance of four cost aggregation techniques, namely box-aggregation using Summed Area Tables, hierarchical aggregation utilizing a pyramidal decomposition, aggregation employing recursive Gaussian filtering, and aggregation exploiting the color characteristics of the images through recursive bilateral filtering. The latter is the most complex one and expected to provide best results. Surprisingly enough, the results from experiments on a number of datasets favor the use of the hierarchical aggregation which shows stability and is also fast and suitable for parallelization on e.g. GPU.

Julkaisufoorumi-taso