Resumo

MARTINEZ, Arley Bejarano; SALCEDO, Andrés Felipe Calvo  e  BAENA, Carlos Alberto Henao. SPACE-FREQUENCY DESCRIPTORS FOR AUTOMATIC IDENTIFICATION OF TEXTURE PATTERNS USING SUPERVISED LEARNING. Cienc. Ing. Neogranad. [online]. 2018, vol.28, n.2, pp.63-82. ISSN 0124-8170.  https://doi.org/10.18359/rcin.3212.

This article presents an evaluation of frequency-space descriptors and texture analysis techniques for textile classification. The work methodology consists of three fundamental stages: characterization, classification and validation. The characterization stage uses descriptors such as wavelet transform, Fourier transform, a state-of-the-art texture characterization method such as segmentation-based fractal texture analysis (SFTA) and the adaptation of the short-space Fourier transform. The classification stage analyzes the use of three state-of-the-art methods such as Artificial Neural Networks (ANN), Support Vector Machines (SVM) and the Gaussian Process (GP); linear, Gaussian and polynomial kernels were included in SVM and GP. To validate the method, an annotated database is built with ten types of fabrics and 1,000 photos, to which the characterization and classification process is applied by means of a Monte Carlo experiment. At this stage, random training (70 %) and testing (30 %) configurations are generated, finding the performance of each classification model. Finally, the confusion matrix is obtained, and the success percentages of each experiment are determined. Additionally, a time analysis is carried out for each algorithm, both at the descriptor and classifier levels, in order to determine the configuration that offers better features and its computational cost.

Palavras-chave : Fourier transform; fractal segmentation; machine learning; textile; texture analysis; time-frequency.

        · resumo em Espanhol     · texto em Espanhol     · Espanhol ( pdf )