Full 3D Compact Histogram Segmentation of Color Images
Résumé
In this study we present a fast unsupervised segmentation method of color images based on the 3D compact histograms. In order to resolve the problems of handling 3D histograms and to palliate the defects of the marginal and bi-marginal approaches, the use of the 3D compact histogram to accomplish classification in the color space RGB proved to be necessary. The interest of this study is based on the use of compact histogram which is a structure with reduced volume of colors without loss of data of the classical histogram and favouring a fast segmentation of color images. Knowing that there is no universal method of segmentation, our aim is to realize an acceptable segmentation that is the most relevant in the sense of a criterion of evaluation of segmentation. The relevance of the choice of peaks is not evident generally and is complexified when the dimension of the space increases; because of the integration of fuzzy neighborhood associated with an algorithm of connected components labeling in our algorithm of classification establishes the first approach of segmentation. The second approach consists in re-quantifying the colors of the histogram before making the classification, this limit the time consuming of segmentation and the effects of over-segmentation; this also the purpose of the first approach. A stage of performance appraisal of segmentation allows bringing to light the relevance of both methods and a comparison of both approaches is so discussed.
Origine : Fichiers produits par l'(les) auteur(s)