Dipartimento d'Ingegneria

Lunedì, 19 Novembre 2018 11:25

Hand-designed descriptors vs. pre-trained convolutional networks: a comparison of two strategies for colour texture classification In evidenza

Scritto da 

ELG 10 at 14:00 Tuesday 20th November 2018.
City, University of London.

Abstract:
Colour texture classification plays a fundamental role in a wide range of applications, as for instance surface inspection and grading, content-based image retrieval, computer-assisted diagnosis and remote sensing. Traditionally, the problem has been approached by manually designing suitable functions capable of extracting meaningful visual features from the input images: the so-called ‘hand-crafted’ paradigm. In recent years, however, the arrival on the scene of Convolutional Neural Networks (`Deep Learning') has brought about dramatic advances in many areas of computer vision, and has the potential to significantly change the approach to texture analysis as well. This seminar will comparatively evaluate – both at a theoretical and experimental level – the hand-crafted and Deep Learning paradigms for colour texture classification. There will be specific focus on the use of pre-trained networks as generic feature extractors for colour textures.


Abstract: Colour texture classification plays a fundamental role in a wide range of applications, as for instance surface inspection and grading, content-based image retrieval, computer-assisted diagnosis and remote sensing. Traditionally, the problem has been approached by manually designing suitable functions capable of extracting meaningful visual features from the input images: the so-called ‘hand crafted’ paradigm. In recent years, however, the arrival on the scene of Convolutional Neural Networks (`Deep Learning') has brought about dramatic advances in many areas of computer vision, and has the potential to significantly change the approach to texture analysis as well. This seminar will comparatively evaluate – both at a theoretical and experimental level – the hand-crafted and Deep Learning paradigms for colour texture classification. There will be specific focus on the use of pre-trained networks as generic feature extractors for colour textures.

Letto 498 volte Ultima modifica il Lunedì, 19 Novembre 2018 11:42

Lascia un commento

Assicurati di inserire (*) le informazioni necessarie ove indicato.
Codice HTML non è permesso.

NOTA:

Questo sito utilizza i cookies, anche di terze parti, per le statistiche e per agevolare la navigazione nelle pagine del sito. Maggiori informazioni disponibili nell’informativa sulla privacy. Per saperne di piu'

Approvo