Akademik Veri Yönetim Sistemi
Anadolu University Journal of Science and Technology, cilt.2, ss.219-225, 2001 (Hakemli Dergi)
Texture classification is an important task in scene analysis, remote sensing, defect recognition from images for quality control and other industrial application areas. Recently, wavelet-based methods e.g. : DWT (Discrete Wavelet Transform), CWT (Continuous Wavelet Transform), Wavelet Packets, Wavelet frames have been proposed for texture features extraction. In this note, a new windowing algorithm is proposed, which forms variable sizes texture sub-images randomly rotated between 0° and 360° for training neural networks classifier with fast adaptive backpropagation algorithm. Non-subsampled wavelet frame transform has been used for feature extraction of 16 textures from a set of Brodatz' album, by means of various wavelet families. Very good classification performance has been obtained with the new windowing technique, when compared with that of the classical non-overlapping windowing method.