Deep Sparse Auto-Encoder Features Learning for Arabic Text Recognition

One of the most recent challenging issues pattern recognition and artificial intelligence is Arabic text recognition. This research topic still a pervasive unaddressed field, because several factors. Complications arise due to cursive nature writing, character similarities, unlimited vocabulary, use multi-size mixed-fonts, etc. To handle these challenges, an automatic requires building robust system by computing discriminative features applying rigorous classifier together achieve improved performance. In this work, we introduce new deep learning based that recognizes contained in images. We propose novel hybrid network, combining Bag-of-Feature (BoF) framework for feature extraction on Sparse Auto-Encoder (SAE), Hidden Markov Models (HMMs), sequence Our proposed system, termed BoF-deep SAE-HMM, tested four datasets, namely printed line images Printed KHATT (P-KHATT), benchmark word Text Image (APTI), handwritten IFN/ENIT, digits Modified National Institute Standards Technology (MNIST).