Yazar "Ezel, Elif" seçeneğine göre listele

Listeleniyor 1 - 1 / 1
  • Küçük Resim
    Öğe
    Derin öğrenme yöntemi kullanılarak görüntü-tabanlı türk işaret dili tanıma
    (Selçuk Üniversitesi, 27.06.2018) Ezel, Elif; Baykan, Ömer Kaan
    Sign language is a form of visual communication that people with hearing problems use to express themselves. The main goal of this study is to facilitate the life of people with hearing problems. In this study, it has been provided to translate the movements belonging to the Turkish Sign Language (TSL) finger alphabet, which is a visual language, into the written language using Convolutional Neural Networks (CNN), one of the deep learning methods. In this study, deep neural networks were used instead of classical machine learning methods and the success of these structures in recognizing the signs of the TSL finger alphabet was evaluated. In this thesis, a dataset was obtained using 522 RGB images taken from three different individuals. For RGB images in the dataset, pre-processing and segmentation were applied and a new dataset consisting of binary images containing only hand regions has been achieved. Data augmentation methods have been applied to the datasets, the number of images increased from 522 to 4176 and it was evaluated how increasing the number of data influences classification success. In order to classify and identify the signs, both the designed 14-layer CNN structure and the AlexNet structure, a pre-trained model, were used. While the binary image dataset was used as the input data for the 14-layer CNN structure, three different dataset consisting of RGB images were used as input data for the AlexNet structure. In addition, Histogram of Oriented Gradients (HoG) feature extraction method was applied to the binary image dataset and the obtained feature vector was used for classification using Artificial Neural Networks (ANN), Support Vector Machines (SVM) and Random Forest (RF). The results were compared and the most successful method was determined.