A comparison of feature detectors and descriptors for object class matching

Antti Hietanen, Jukka Lankinen, Joni-Kristian Kämäräinen, Anders Glent Buch, Norbert Krüger

Publikation: Bidrag til tidsskriftKonferenceartikelForskningpeer review

Resumé

Solid protocols to benchmark local feature detectors and descriptors were introduced by Mikolajczyk et al. 1 and 2. The detectors and the descriptors are popular tools in object class matching, but the wide baseline setting in the benchmarks does not correspond to class-level matching where appearance variation can be large. We extend the benchmarks to the class matching setting and evaluate state-of-the-art detectors and descriptors with Caltech and ImageNet classes. Our experiments provide important findings with regard to object class matching: (1) the original SIFT is still the best descriptor; (2) dense sampling outperforms interest point detectors with a clear margin; (3) detectors perform moderately well, but descriptors׳ performance collapses; (4) using multiple, even a few, best matches instead of the single best has significant effect on the performance; (5) object pose variation degrades dense sampling performance while the best detector (Hessian-affine) is unaffected. The performance of the best detector-descriptor pair is verified in the application of unsupervised visual class alignment where state-of-the-art results are achieved. The findings help to improve the existing detectors and descriptors for which the framework provides an automatic validation tool.
OriginalsprogEngelsk
TidsskriftNeurocomputing
Vol/bind184
Udgave nummerC
Sider (fra-til)3-12
ISSN0925-2312
DOI
StatusUdgivet - 2016
BegivenhedRobust Local Descriptors for Computer Vision - , Singapore
Varighed: 1. nov. 20145. nov. 2016

Konference

KonferenceRobust Local Descriptors for Computer Vision
LandSingapore
Periode01/11/201405/11/2016

Fingeraftryk

Benchmarking
Detectors
Sampling

Citer dette

Hietanen, Antti ; Lankinen, Jukka ; Kämäräinen, Joni-Kristian ; Buch, Anders Glent ; Krüger, Norbert. / A comparison of feature detectors and descriptors for object class matching. I: Neurocomputing. 2016 ; Bind 184, Nr. C. s. 3-12.
@inproceedings{9c170613033e415c9f0bd81a546cbd2a,
title = "A comparison of feature detectors and descriptors for object class matching",
abstract = "Solid protocols to benchmark local feature detectors and descriptors were introduced by Mikolajczyk et al. 1 and 2. The detectors and the descriptors are popular tools in object class matching, but the wide baseline setting in the benchmarks does not correspond to class-level matching where appearance variation can be large. We extend the benchmarks to the class matching setting and evaluate state-of-the-art detectors and descriptors with Caltech and ImageNet classes. Our experiments provide important findings with regard to object class matching: (1) the original SIFT is still the best descriptor; (2) dense sampling outperforms interest point detectors with a clear margin; (3) detectors perform moderately well, but descriptors׳ performance collapses; (4) using multiple, even a few, best matches instead of the single best has significant effect on the performance; (5) object pose variation degrades dense sampling performance while the best detector (Hessian-affine) is unaffected. The performance of the best detector-descriptor pair is verified in the application of unsupervised visual class alignment where state-of-the-art results are achieved. The findings help to improve the existing detectors and descriptors for which the framework provides an automatic validation tool.",
author = "Antti Hietanen and Jukka Lankinen and Joni-Kristian K{\"a}m{\"a}r{\"a}inen and Buch, {Anders Glent} and Norbert Kr{\"u}ger",
year = "2016",
doi = "10.1016/j.neucom.2015.08.106",
language = "English",
volume = "184",
pages = "3--12",
journal = "Neurocomputing",
issn = "0925-2312",
publisher = "Elsevier",
number = "C",

}

A comparison of feature detectors and descriptors for object class matching. / Hietanen, Antti; Lankinen, Jukka; Kämäräinen, Joni-Kristian; Buch, Anders Glent; Krüger, Norbert.

I: Neurocomputing, Bind 184, Nr. C, 2016, s. 3-12.

Publikation: Bidrag til tidsskriftKonferenceartikelForskningpeer review

TY - GEN

T1 - A comparison of feature detectors and descriptors for object class matching

AU - Hietanen, Antti

AU - Lankinen, Jukka

AU - Kämäräinen, Joni-Kristian

AU - Buch, Anders Glent

AU - Krüger, Norbert

PY - 2016

Y1 - 2016

N2 - Solid protocols to benchmark local feature detectors and descriptors were introduced by Mikolajczyk et al. 1 and 2. The detectors and the descriptors are popular tools in object class matching, but the wide baseline setting in the benchmarks does not correspond to class-level matching where appearance variation can be large. We extend the benchmarks to the class matching setting and evaluate state-of-the-art detectors and descriptors with Caltech and ImageNet classes. Our experiments provide important findings with regard to object class matching: (1) the original SIFT is still the best descriptor; (2) dense sampling outperforms interest point detectors with a clear margin; (3) detectors perform moderately well, but descriptors׳ performance collapses; (4) using multiple, even a few, best matches instead of the single best has significant effect on the performance; (5) object pose variation degrades dense sampling performance while the best detector (Hessian-affine) is unaffected. The performance of the best detector-descriptor pair is verified in the application of unsupervised visual class alignment where state-of-the-art results are achieved. The findings help to improve the existing detectors and descriptors for which the framework provides an automatic validation tool.

AB - Solid protocols to benchmark local feature detectors and descriptors were introduced by Mikolajczyk et al. 1 and 2. The detectors and the descriptors are popular tools in object class matching, but the wide baseline setting in the benchmarks does not correspond to class-level matching where appearance variation can be large. We extend the benchmarks to the class matching setting and evaluate state-of-the-art detectors and descriptors with Caltech and ImageNet classes. Our experiments provide important findings with regard to object class matching: (1) the original SIFT is still the best descriptor; (2) dense sampling outperforms interest point detectors with a clear margin; (3) detectors perform moderately well, but descriptors׳ performance collapses; (4) using multiple, even a few, best matches instead of the single best has significant effect on the performance; (5) object pose variation degrades dense sampling performance while the best detector (Hessian-affine) is unaffected. The performance of the best detector-descriptor pair is verified in the application of unsupervised visual class alignment where state-of-the-art results are achieved. The findings help to improve the existing detectors and descriptors for which the framework provides an automatic validation tool.

U2 - 10.1016/j.neucom.2015.08.106

DO - 10.1016/j.neucom.2015.08.106

M3 - Conference article

VL - 184

SP - 3

EP - 12

JO - Neurocomputing

JF - Neurocomputing

SN - 0925-2312

IS - C

ER -