Aplicação dos Algoritmos SIFT e SURF na Classificação de Sub-Imagens por Discriminação de Textura

Image segmentation is present in the development of many technologies and applications. The proposal of this work is to evaluate and compare the performance of SIFT and SURF algorithms in a classification process of sub-images, where this classification is done considering the different textures in the image. For the purpose of this work, the textures of the image may be natural, with random and complex behaviors, or being formed by similar objects of complex nature. This work may be useful to assist in image segmentation processes. In the development of this work, artificial neural networks are used to classify the descriptors vectors generated with SIFT and SURF algorithms, when applied on different sub-images. The neural networks used are of SOM type, therefore use a non-supervised learning for training. The training of the neural networks is done with input patterns generated through SIFT and SURF algorithms when applied on some sub-images. To carry out the work studies were done on image segmentation methods, artificial neural networks and the functioning of SIFT and SURF algorithms.