Digits Recognition for Arabic Handwritten through Convolutional Neural Networks, Local Binary Patterns, and Histogram of Oriented Gradients

Abstract

The recognition of handwritten text is a topic of study that has several applications. One of these applications is the recognition of handwriting in official documents, historical scripts, bank checks, etc., which is a problem that might be considered relatively a security issue. The topic of handwriting recognition has been the subject of a significant amount of study and analysis in recent years. People from a variety of countries, including all of the countries that use Arabic as their primary language, as well as Persian, Urdu, and Pashto languages, also use Arabic characters in their scripts. As people's handwriting is infinitely varied, recognition systems confront numerous challenges. This paper aims to examine the efficacy of some techniques in addressing the problem of Arabic Handwritten Numbers Recognition (AHNR). Specifically, the methods under consideration are Convolutional Neural Networks (CNNs), which have demonstrated their utility in diverse domains and offer effective solutions. Local Binary Pattern (LBP) is a unique, efficient textural operator that finds widespread application in the area of computers such as biometric identification and detection of targets as feature extraction techniques. In addition, a Histogram of Oriented Gradients (HOG) is a feature extraction technique that is used in computer vision and image processing for the purpose of object detection. The HOG descriptor focuses on the structure or the shape of an object. It is better than any edge descriptor as it uses magnitude as well as the angle of the gradient to compute the features. Furthermore, the K-Nearest Neighbor (KNN) algorithm will be employed as a classifier in conjunction with LBP and HOG. Comparing the performance of the three methods, the (CNN) model achieved nearly 99% recognition accuracy, which is asymptotic for the HOG approach. In terms of computational efficacy, the CNN model was 0.61 seconds faster than the HOG approach.

Keywords

• Image Processing, Deep Learning, K Nearest Neighbor, Pattern Recognition, Machine Learning