Text classification by machine learning algorithms using a new text feature extraction method based on image processing

Accurate text and character identification on documents using smart technologies is a very important method of obtaining data. The complex and irregular text and characters on the images, as well as the use of different writing styles, affect the text recognition success of both Artificial Intelligence (AI) and Machine Learning (ML) technologies. Manually transferring texts and characters from paper format documents to digital media creates a great waste of time and labor. In addition, when documents containing direct text are scanned and transferred in a computer environment, the texts cannot be edited. OCR (Optical Character Recognition) methods, which are proposed as a solution to this situation, are one of the Natural Language Processing (NLP) tasks. In particular, it has been observed that even in current artificial intelligence-based OCR software, the characters 0 and O are confused with each other. In this study, it is suggested that image pre-processing should be done on images containing characters in order to increase the success of character recognition. In the study, a new model was designed to increase the success of correctly recognizing 0 and O characters that are very similar to each other. In the study, image pre-processing was applied to the images of 408 characters. Classification successes were measured by using kNN, SVM and Logistic Regression algorithms on the data set. Additionally, the classification performance of 0 and O characters was measured on the artificial intelligence-based Google Documents tool. According to the results obtained, the success of recognizing 0 and O characters with the LR machine learning algorithm was realized at the rate of 1.00 according to the performance metrics.