Deep Learning Autoencoder Approach for Handwritten Arabic Digits Recognition

Abstract: This paper presents a new unsupervised learning approach with stacked autoencoder (SAE) for Arabic handwritten digits categorization. Recently, Arabic handwritten digits recognition has been an important area due to its applications in several fields. This work is focusing on the recognition part of handwritten Arabic digits recognition that face several challenges, including the unlimited variation in human handwriting and the large public databases. Arabic digits contains ten numbers that were descended from the Indian digits system. Stacked autoencoder (SAE) tested and trained the MADBase database (Arabic handwritten digits images) that contain 10000 testing images and 60000 training images. We show that the use of SAE leads to significant improvements across different machine-learning classification algorithms. SAE is giving an average accuracy of 98.5%.

Keywords: Deep Learning, Stacked autoencoder, Arabic Digits recognition

1. Introduction Pattern recognition has become a massive important due to ever demanding need of machine learning and artificial intelligence in practical problems [1]. Handwritten digits recognition is one such problem in which digits written by different authors are recognized by machines [2]. Recognition digits covers many applications such as office automation, check verification, postal address reading and printed postal codes and data entry applications are few applications [3]. Deep learning (DL) is a hierarchical structure network which through simulates the human brain’s structure to extract the internal and external input data’s features [4]. Deep learning based on algorithms using multilayer network such as deep neural networks, convolutional deep neural networks, deep belief networks, recurrent neural networks and stacked autoencoders. These algorithms allow computers and machines to model our world well enough to exhibit the intelligence. Autoencoders is an artificial neural network used for learning efficient encoding where the input layer have the same number of the output layer where the hidden layer has a smaller size [5,6]. In the autoencoder, the hidden layer gives a better representation of the input than the original raw input, and the hidden layer is always the compression of the input data which is the important features of the input. So, the propose of the paper is using Stacked Auto- Encoder (SAE) to create deep learning recognition system for Arabic handwritten digits recognition. The rest of the paper is organized as follows: Section 2 gives a review on some of the related work done in the area. Section 3 describes the proposed approach, Section 4 gives an experimental result, and we list our conclusions and future work in Section 5.
2. Related Work Various methods for the recognition of Latin handwritten digits [1,7,8,9] have been proposed and high recognition rates are reported. On the other hand,
   many researchers addressed the recognition of digits including Arabic.
   In 2008, Mahmoud [11] presented a method for the automatic recognition of Arabic handwritten digits using Gabor-based features and Support Vector Machines (SVMs). He used a medium database have 21120 samples written by 44 writers. The database contain 30% for testing and the remaining 70% of the data is used for training. They achieved average recognition rates are 99.85% and 97.94% using 3 scales & 5 orientations and using 4 scales & 6 orientations, respectively.
   In 2011, Melhaoui et al. [10] presented an improved technique based on Loci characteristic for recognizing Arabic digits. Their work is based on handwritten and printed numeral recognition. They trained there algorithm on dataset contain 600 Arabic digits with 400 training images and 200 testing images. They were able to achieve 99% recognition rate on this small database. In 2013, Pandi selvi and Meyyappan [2] proposed an approach to recognize handwritten Arabic numerals using back propagation neural network. The final result shows that the proposed method provides a

recognition accuracy of more than 96% for a small sample handwritten database. In 2014, Takruri et al. [12] proposed three level classifier based on Support Vector Machine, Fuzzy C Means and Unique Pixels for the classification of handwritten Arabic digits. they tested the new algorithm on a public dataset. The dataset contain 3510 images with 40% are used for testing and 60% of images are used for training. The overall testing accuracy reported is 88%. In 2014, Majdi Sala

…(Full text truncated)…