Mojdeh Vafaei - Department of Electrical Engineering Najafabad Branch, Islamic Azad University
Hossein Pourghassem - Department of Electrical Engineering, Najafabad Branch, Islamic Azad University,

Carpet map image classification and retrieval bycomputer are important issues in Persian carpet industry. In thispaper, a novel carpet map classification algorithm based ondirectional and texture features is proposed. The carpet mapimages are classified into two major classes (Kheshti map classand Lachak-Toranj map class) which have directional andtextural distinctions. For this purpose, Gabor filter, Wavelettransform and intensity histogram are used to extract feature.Support Vector Machine and Multilayer Perceptron are appliedto classify. The proposed algorithm is evaluated on a set of carpetmap images that had been captured from real carpets. Theobtained results by our algorithm demonstrate the efficiency thisalgorithm in content-based image retrieval applications.

Carpet map classification; Gabor filter; Wavelet transform; Multilayer Perceptron; Support Vector Machine