Grammatical Inference and SIFT for Scene Recognition RIBAS, L. C. ; BORTH, M. ; CASTRO JUNIOR, A. A. ; GONÇALVES, W. N. ; PISTORI, H. X Workshop de Visão Computacional, 2014. p. 293-298.
Grammatical inference in computer vision regained attention in recent years due to the emergence of new local feature techniques, such as Scale Invariant Feature Transform (SIFT) e Speeded Up Robust Features (SURF). This paper presents a methodology that converts an image into a string based on the SIFT and bag-of-visual-words (BOW). Given the strings, grammar induction techniques are used for scene recognition. In the BOW, the vocabulary is usually learned using an unsupervised approach. To improve image description, this paper proposes a supervised vocabulary learning. This approach improves the string obtained from the image and the experimental results have demonstrated its robustness in a challenging scene database.