Letter Recognition Using Machine Learning Algorithms
Optical character recognition represents the mechanical or electronic conversion of handwritten, typed or printed images into coded text. Optical character recognition is widely used as a form of data entry from records that have been printed, and it can include invoices, bank statements, passports and many more. In the research, Optical character recognition reads data from the Re-Captcha dataset of images, converts them into strings, and these strings are used for testing, training and calculating prediction accuracy. The methodologies used are Convolutional neural network and Recurrent neural network. The convolutional neural network consist of neurons that receive data and group them according to similarity. A recurrent neural network cycle can be created between the connections of nodes, allowing the output from nodes to influence the subsequent input to other nodes. For data were used Re-Captcha images, and for the prediction of characters from images was used TensorFlow with Keras. The best results that are produced can be compared between first and last result, where the loss for first result was 20.63 and value loss was 16.45, while last result has loss of 0.56 and value loss of 2.96