Text Detection and Recognition Using Fusion Neural Network Architecture
Sunil Kumar Dasari, Shilpa Mehta · 2022 8th International Conference on Advanced Computing and Communication Systems (ICACCS) · 2022
Text recognition possesses various application which attracted wide number of research interests in past decade. Recent method includes the deep learning-based model and it follows the attention-based mechanism. However, this deep learning mechanism suffers problem with recurrence alignment and also the feature extraction and labelling sequence becomes hard due to involvement of image as well as text data. Hence, this research designs and develops a deep learning-based Fusion Neural Network for detection and recognition of text. FNN aims to improve the accuracy of identification. Fusion Neural Network comprises different layer of different Neural Network; moreover, convolutional layer is used to obtain the feature sequence and optimal training model is designed using recurrent layer to enhance the accuracy. FNN is evaluated considering Devanagari MLT-19 dataset, evaluation is carried out considering different parameter, at first script cropping and recognising is carried out, also comparison with existing methodology is carried out to prove proposed model efficiency. Furthermore, FNN model observes 98.67% of script word identification accuracy.