Hybrid and incremental fuzzy learning to classify pixels by colors. / Aprendizado nebuloso híbrido e incremental para classificar pixels por cores.

AUTOR(ES)

Waldemar Bonventi Junior

DATA DE PUBLICAÇÃO

2005

RESUMO

Image segmentation is a very important process, which aims at subdividing an image in parts that correspond to objects of interest in the application domain. Objects may depict few colors that are represented in an image by a set of pixels presenting a very large range of chromatic values. A relatively small number of human-defined linguistic labels can be assigned to these sets, and these labels characterize colors represented by classes. However, the borders among these classes are fuzzy, since the chromatic values that define the transition from a class to another depend on different domain factors. This thesis contributes in the image segmentation process by proposing a pixel classifier based exclusively on the color attribute. Fuzzy sets theory is used to deal with the problem of fuzziness among color classes. This thesis proposes a hybrid and incremental scheme for learning fuzzy models of color classes included in the classifier. The hybrid-learning scheme combines unsupervised and supervised learning paradigms, transferring the labeling by a supervisor from individual instances (a very computationally costly task) to groups of similar instances. These groups are combined by application of adequate aggregation operators, providing an incremental learning scheme to the classifier, so that models can be revised and new classes can be incorporated into the models. In order to provide completeness to the models, a generalization process is also proposed. The classifier was tested in the human skin color-modeling problem, by using digital color-images captured under controlled and uncontrolled conditions. Experimental results assess its effectiveness, providing a robust and accurate detection of skin color in digital color-images.

ASSUNTO(S)

image processing fuzzy (inteligência artificial) processamento de imagens aprendizado computacional fuzzy sets machine learning

Documentos Relacionados

• Modeling and predictive hybrid control by fuzzy logic for polymerization processes
• Desenvolvimento e análise de controle híbrido preditivo por lógica fuzzy de processos de polimerização
• HIERARQUICAL NEURO-FUZZY MODELS BASED ON REINFORCEMENT LEARNING FOR INTELLIGENT AGENTS
• Contributions to projection pursuit learning
• Purification and characterization of bovine rotavirus cores.