Current Proceedings on Technology
Yazarlar: Iclal Ulvi, Rifat Asliyan, Korhan Gunel, Onur Dincel, Aslihan Cagrici
Konular:-
Anahtar Kelimeler:Image classification,Kernel based categorization,Textile images,Artificial neural networks,MLP,SOFM,LVQ
Özet: Rapid developments in textile industry require the categorization of textile motifs because of great increase in the number of textile images. In this paper, the systems, which automatically detect textile motifs in images, have been developed. In general, the developed systems consist of the stages of preprocessing, feature extraction, training and testing. Six edge detection and skeletonization operations have been applied to all textile images. In feature extraction, each image has been represented as a vector including the probabilities of predefined 2x2 and 3x3 kernels in the image. In training stage, the images are classified into seven categories using MLP, LVQ and SOFM methods. According to the accuracy and F-measure values, the results of the systems have been evaluated by utilizing the images in the test set. In this study, SOFM is the most successful method according to accuracy and F-measure scores as 96.2% and 86.9%, respectively.