Caracterización y clasificación de glóbulos blancos mediante descriptores locales de imágenes

Abstract

The purpose of this work is to other a study of the analysis, exploration and comparison of a series of extractors of local features or points of interest of different types, in order to be part of the description of required attribute vectors in a classi- fication scheme later denominated Bag of Visual Words (Visual BoW). It is a model of learning used in a wide range of disciplines, but unexplored for this problem that arises this work: automatic classification of color images with examples of white blood cells of different classes. Therefore, several feature detectors allow us to know the type of detected feature available that best ts the problematic raised, improving the accuracy and robustness of the chosen classi cation method. In addition to this study, other phases of the Visual BoW process are also discussed, such as the choice of parameters in the process of "vocabulary" formation and the supervised learning algorithm.