부산대학교 도서관

Aiming at optical mark reader (OMR) recognization problems and the complexity of virtual operating that has resulted in much increased human labor, the model architecture based on convolution neural network(CNN) and TensorFlow platform is proposed. In order to improve OMR recognition technology, we uses deep learning tools in TensorFlow to construct the neural network model. For the sake of training the model we uses TensorFlow to preprocess the OMR input data, obtains the TFRecord file, and then train model to reach the requisite OMR recognition level in test paper image and obtain the standard parameters. The view tool TensorBoard is used to display the calculated flow graph of the TensorFlow model.