Makİne Öğrenmesİ yÖntemlerİ İle tomatİk ÇevrİmdiŞi İmza tanima ve doğrulama Sistemİ

In this work, a low-cost and efficient, offline signature verification system was developed. The system is designed to be composed of 4 steps: data collection, preprocessing, feature extraction and signature verification. The pre-processing phase is completed in 4 steps: graying, normalization, thresholding and image skeleton. Normalization is performed by nearest neighbor interpolation. The thresholding step is performed with the Otsu algorithm. In the last stage; The skeleton of the image is performed using the Zhang algorithm. In the third stage of the system, the feature extraction step, the Hu's invariant moments, regional features and the discrete wavelet transform were used. The generated data set has been tested separately for the selected training algorithms and for signature recognition and signature verification. As a result of the observations made, it is observed that the SVM algorithm gave the best result for signature verification.