Transferring attendance lists to the digital platform with computer vision and detection of fraud in signature

In this thesis, two methods are proposed for transferring attendance lists to the digital spreadsheets using computer vision methods. The number of attendances of students is calculated automatically by using the proposed methods. In the preprocessing step, the attendance paper in the photograph and signature cells on the attendance paper are detected by using image processing methods. The first method proposed for signature detection is "The Edge and Skeleton Method". In this method, horizontal and vertical pixel numbers of the edges and skeletal structures of the cells are used in the developed algorithm to conclude whether the cell contains a signature or not. The second method is developed by using 64 and 128 filtered Convolutional Neural Networks (CNN). The trained CNNs predict the presence of the signatures in the cells. The prediction accuracies of the proposed methods for signature detection on the attendance papers are as follows: The Edge and Skeleton Method is 97.61%, CNN with 64 filters is 98.39% and CNN with 128 filters is 98.48%. The corresponding execution times of the methods are 64.24, 82.07, and 92.81 seconds. Thus, while the prediction ratios of the methods are very proximate, The Edge and Skeleton method outperforms the CNN based methods by 27.76% and 44.47% in terms of execution time. Additionally, one more method is proposed for signature fraud on the attendance paper. The method is developed using CNN and informs for possible signature fraud with a success ratio of 89.47%. Key Words: Computer vision, Convolutional neural networks, Optical character recognition, Image processing, Shape skeleton, Canny edge detection 2021, 94 pages