Convolution Neural Networks for Arabic Font Recognition

Abstract

Designers have a large number of fonts to choose from. But what if they see the perfect font but do not know its name? In this paper we present a smart font recognition system, that takes a picture of the desired font, and helps with font identification. In this paper, we explore the usage of convolution neural networks to find the name of a font, given its picture. To achieve this task, a new Arabic font dataset is created. The dataset consists of 2500 pictures of single words covering 50 Arabic fonts. This dataset is then used to train different deep neural networks such as AlexNet, ResNet and other architectures to recognize single word fonts. Finally the results of the different models are compared and the best model was implemented using the sliding window technique in order to classify a full paper containing a whole text.