The Effectiveness Of Transfer Learning For Arabic Handwriting Recognition Using Deep Cnn

Recognition of handwritten words is a widely used system in the modern world that remains a significant challenge. Traditional machine learning techniques require creative engineering and considerable domain expertise to transform the raw data into a feature vector from which the classifier could rank the input model. To deal with this problem, the popular convolutional neural networks (CNNs), and Its Derivatives introduced recently, have effectively replaced handcrafted descriptors with network features and have been found to provide significantly better results than traditional methods. It is one of the fastest growing fields in machine learning, promising to reshape the future of artificial intelligence. However, the problem with deep learning is that it requires large data sets for training due to the large number of parameters needed to be tuned by a learning algorithm. In our proposed approach, Deep Convolutional Neural Networks (DCNN) is adapted to perform the classification phase. Notably the architectures Inception-v3, ResNet, and VGG16, using an enriched dataset containing images extracted from the IFN/ ENIT database. To refine pre-trained models for deep characteristics extraction, we utilize Transfer Learning technique. This technique has shown very good performance for the frequent recognition problems. According to the results obtained, the developed system gives very interesting recognition rates.