Wyniki wyszukiwania - BazTech

Ograniczanie wyników

1 Machine Graphics and Vision

1 2009

Znaleziono wyników: 1

Liczba wyników na stronie

Wyniki wyszukiwania

Wyszukiwano:
w słowach kluczowych: clustering stability

Sortuj według:

Ogranicz wyniki do:

Clustering Stability-Based Feature Selection for Unsupervised Texture Classification

Klepaczko A., Materka A.

Machine Graphics and Vision

2009

Vol. 18, No. 2

125-141

This paper addresses the issue of unsupervised texture image classification. This kind of analysis can augment automatic image interpretation and recognition whenever visualized objects surfaces reveal some regular shapes or patterns. The variety of texture models and parameters calculated therein implies the need to select the relevant attributes which allow the best possible texture discrimination. However, it has been observed that the discriminative power of the texture parameters deteriorates if the image dimensions are small relative to the size of a single texture element. In that case, feature vectors corresponding to different textures become less distinguishable. Although this does not constitute a significant impediment to supervised feature selection, the methods which operate in an unsupervised manner and are analyzed in this study perform well only if the images being classified contain a large portion of texture. We illustrate this phenomenon through a series of experiments with natural, Brodatz-album texture images. We also show how to overcome the outlined problem by assessing features saliency using a measure based on the notion of clustering stability.