Estudo e implementação de técnicas para reconhecimento de objetos em imagens.

Abstract

One of the most used step in handling and processing data in intelligent systems is to recognize objects. Therefore, the study of techniques for recognizing objects is extremely important, since there is a need to develop effective systems with high recognition rates, which are invariant to rotation, noise, light, etc. Such task can be defined as follow: An identifying activity of an individual object as a member of a particular class that contains objects with similar characteristics. To implement that algorithm, descriptors and classifiers are used. In this context, this work involved an initial study of object recognition techniques, towards the implementation of one of its possible algorithms, in order to valid an identification system. The implemented algorithm used HOG as the descriptor and SVM as the classifier, as well, the system was subjected to three different databases being evaluated the recognition rate for each situation. The databases used were: Caltech-101, MSRC v1 and Stanford Cars Dataset.