Use of Discrete Cosine Transformation and Histograms of Oriented Gradients for Optical Arabic Word Recognition of Different Font Styles and Sizes
For our proposal, we used discrete cosine transform descriptors (DCT) and the histogram of oriented gradients (HOG) for word-level recognition. Extracting and classifying features from word-images are features of all machine learning algorithms. We used the k-Nearest-Neighbour (KNN) as classifiers and evaluated the features and classifiers on seven fonts.
Four fonts, specifically Times New Roman, Arial, Arabic Transparent and Simplified Arabic are used for written books, magazines, and research papers. Three other fonts, namely Arial Unicode MS, Tahoma and K Traffic, are used for adverts and CAPTCHA. Each font type includes different sizes (20, 24, and 28). We obtained high accuracy in the proposed features using the KNN.
The results show that the features significantly outperformed by using the pixel density directly from the images.
Keywords: Optical Character Recognition (OCR); Features Descriptors; Features Extraction; DCT; HOG; Features Vectors, KNN
Download Full-Text
ABOUT THE AUTHORS
Naser Telesi
Has received the B.Sc. at the Faculty of Computer Science, Tripoli University, Tripoli-Libya. MSc at Libyan Academia form 2007. Now, Ph.D. student. major research focuses on the Pattern Recognition, Arabic character recognition
Mohamed Hashmi
Has received his B.Sc. at the Faculty of the Computer Science, Tripoli University, Tripoli-Libya, 1990. M.Sc, Electronic Design Automation, Beijing University of Aeronautics & Astronautics, China, 2001, major research focuses on Software Engineering and internet security
Aleksandar Jevremović
PhD, is a full professor at the Singidunum University. So far, he has authored/co-authored number of research papers and made contributions to three books about computer networks, computer networks security and Web development. He is recognized as an Expert Level Instructor at Cisco Networking Academy program.
Naser Telesi
Has received the B.Sc. at the Faculty of Computer Science, Tripoli University, Tripoli-Libya. MSc at Libyan Academia form 2007. Now, Ph.D. student. major research focuses on the Pattern Recognition, Arabic character recognition
Mohamed Hashmi
Has received his B.Sc. at the Faculty of the Computer Science, Tripoli University, Tripoli-Libya, 1990. M.Sc, Electronic Design Automation, Beijing University of Aeronautics & Astronautics, China, 2001, major research focuses on Software Engineering and internet security
Aleksandar Jevremović
PhD, is a full professor at the Singidunum University. So far, he has authored/co-authored number of research papers and made contributions to three books about computer networks, computer networks security and Web development. He is recognized as an Expert Level Instructor at Cisco Networking Academy program.