Pattern recognition applications for image classification

Abstract

Pattern recognition and image classification applications are the topic of this thesis. Its focus is on neural network based techniques, in other words, convolutional neural networks, and they are tried to be used in the most efficient way. In addition to convolutional neural networks, some experiments using more classical pattern recognition and classi cation techniques are conducted and their results are evaluated. The applications are made for traffic sign recognition and medical image recognition using different public datasets that were also preferred in previous studies. In the context of these applications, the obtained results are also compared to the results available in the literature. As an additional experiment, an optical character recognition and a real text digitalization method is also proposed as a proof of concept study. These experiments again consist of similar steps, but, also a character segmentation approach is developed on real text images, to extract the characters to be classiffied. During all these experiments, image processing and pattern recognition and classification techniques are used, and some proposal are brought to augment their efficiency.