Handwritten Text Recognition Using CRNN
Ahmed A. Idris, Dujan B. Taha · 2022
Text recognition is one of the significant and demanding jobs that needed to keep diving into finding the stability result because of the wide range of real-world applications' use. Individual characters are often detected and recognized independently using traditional methods. Character detection is commonly achieved through the use of sliding windows or related components. The performance of these kinds of algorithms does not work well when the quality of the source image is poor or the image is complex. Convolutional Recurrent Neural Network (CRNN) is a deep learning-based end-to-end text recognition system applied in this study to recognize indefinite-length text sequences. The study proposes CRNN for text recognition where the study will detect the line of text with curvature words by detecting the word area and then will use it to recognize the words one by one by a model trained using CRNN with the IAM dataset words. Accordingly, the Character Error Rate (CER) reached 4.57.