Reconhecimento de padrões por meio de floresta de caminhos ótimos

Abstract

Currently there are large databases available, due to the advances of the acquisition and storage of this information technologies. However, there is a large amount of unlabeled data in relation to a small section labeled. Becoming necessary effective and efficient learning techniques for manipulation and analysis of this information. For learning the recognition of certain patterns is needed, which can be obtained by imaging descriptors, extracting visual properties related to color, form and texture. Some of the images extracted features may be redundant, others are more relevant to the discrimination of the images. Therefore, after the extraction of the characteristics of images, it is important to analyze and obtain the feature vector that best describes the data set by applying dimensional reduction, optimization and normalization techniques. Then, different procedures may be used (supervised, semi-unsupervised and supervised) learning. This work aims to study and the analysis of more effective and efficient techniques for description and classification of bioimages.