In this paper we present an analysis of image featuresused to discriminate arteries and veins in digital fundus im-ages. Methods proposed in the literature to analyze the vas-culature of the retina and compute diagnostic indicators likethe Arteriolar to Venular ratio (AVR), use, in fact, differentapproaches for this classification task, extracting differentcolor features and exploiting different additional informa-tion. We concentrate our analysis on finding optimal fea-tures for the vessel classification, considering not only sim-ple color features, but also spatial location and vessel sizeand testing different supervised labeling approaches. Theresults obtained show that best results are obtained mixingfeatures related with color values and contrast inside andoutside the vessels and positional information. Further-more, the discriminative power of the features changes withthe image resolution and best results are not obtained at thefinest one. Our experiments demonstrate that using a goodset of descriptors it is possible to achieve very good classi-fication performances even without using vascular connec-tivity information.
Effective features for artery-vein classification in digital fundus images
GIACHETTI, Andrea;
2012-01-01
Abstract
In this paper we present an analysis of image featuresused to discriminate arteries and veins in digital fundus im-ages. Methods proposed in the literature to analyze the vas-culature of the retina and compute diagnostic indicators likethe Arteriolar to Venular ratio (AVR), use, in fact, differentapproaches for this classification task, extracting differentcolor features and exploiting different additional informa-tion. We concentrate our analysis on finding optimal fea-tures for the vessel classification, considering not only sim-ple color features, but also spatial location and vessel sizeand testing different supervised labeling approaches. Theresults obtained show that best results are obtained mixingfeatures related with color values and contrast inside andoutside the vessels and positional information. Further-more, the discriminative power of the features changes withthe image resolution and best results are not obtained at thefinest one. Our experiments demonstrate that using a goodset of descriptors it is possible to achieve very good classi-fication performances even without using vascular connec-tivity information.File | Dimensione | Formato | |
---|---|---|---|
paper.pdf
accesso aperto
Tipologia:
Documento in Pre-print
Licenza:
Accesso ristretto
Dimensione
740.89 kB
Formato
Adobe PDF
|
740.89 kB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.