Text Detection and Recognition Based onBidirectional High Order Recurrent Neural Networkand Deep Neural Network

Deep learning is a part of artificial intelligence that allows the computer to learn new rules. In recent years, text recognitionhas inspired many researchers, and still, it needs to improve because of the poor performance of recognition algorithms. This paperproposed a novel method for text recognition that integrates a Bidirectional High-Order Recurrent Neural Network (BHORNN) and adeep Conventional Neural Network (CNN).BHORNN uses more memory units to track previous hidden states, all of which are returned to the hidden layers as feedback throughvarious weight paths. In the proposed method, we should swap the height and width axes, together. (Because after the preprocessing byCNN, when we want to send the picture to a BHORNN, we want the width-axes to be the time-axes because the height size is larger thanthe width), In Optical Character Recognition (OCR), this temporal aspect of a BHORNN allows it to take slices of the image acrossvariable-width characters and recognize it. after that, we pass each of the pixels of the width of the picture that is divided into somepixels, from a patch of CNN. Then we send the picture to a BHORNN and at the end, the BHORNN sends the picture to CTC Loss forevaluation of the performance.

کلیدواژه ها:

Text Detection ، Conventional Neural Network ، OCR ، Natural Language Processing ، Deep Learning