TY - GEN

T1 - Offline Writer Identification using Deep Convolution Neural Network

AU - Durou, Amal

AU - Aref, Ibrahim

AU - Erateb, Suleiman

AU - El-Mihoub, Tarek

AU - Ghalut, Tarik

AU - Emhemmed, Adel

N1 - Conference code: two

PY - 2022/7/27

Y1 - 2022/7/27

N2 - Deep convolutional neural networks (DCNN) are efficient in solving different pattern recognition problems and have been applied to extract image features (IFs). This paper investigates using deep learning (DL) techniques to improve the performance of the writer identification (WI) process. This work presents a novel approach for WI tasks by combining a DL technique with machine learning (ML). A convolutional neural network (CNN) is employed as a feature extractor along with a ML algorithm to classify those features. The standard Alex-Net model is utilized to extract IFs that located in the fully connected layers (FCLs). The support vector machine (SVM) model is selected as the classifier due to its efficient capabilities to improve identification performance (IP). The proposed model is tested using various types of the datasets, namely the Islamic Heritage Project (IHP) and Clusius. Furthermore, IAM and ICFHR-2012 datasets have been employed for benchmarking the proposed model. The results demonstrate the model achieves superior performance.

AB - Deep convolutional neural networks (DCNN) are efficient in solving different pattern recognition problems and have been applied to extract image features (IFs). This paper investigates using deep learning (DL) techniques to improve the performance of the writer identification (WI) process. This work presents a novel approach for WI tasks by combining a DL technique with machine learning (ML). A convolutional neural network (CNN) is employed as a feature extractor along with a ML algorithm to classify those features. The standard Alex-Net model is utilized to extract IFs that located in the fully connected layers (FCLs). The support vector machine (SVM) model is selected as the classifier due to its efficient capabilities to improve identification performance (IP). The proposed model is tested using various types of the datasets, namely the Islamic Heritage Project (IHP) and Clusius. Furthermore, IAM and ICFHR-2012 datasets have been employed for benchmarking the proposed model. The results demonstrate the model achieves superior performance.

KW - Deep convolutional neural network, Writer identification, Feature extraction, Machine learning

U2 - 10.1109/MI-STA54861.2022.9837764

DO - 10.1109/MI-STA54861.2022.9837764

M3 - Conference contribution/Paper

SN - 9781665479196

SP - 43

EP - 47

BT - 2022 IEEE 2nd International Maghreb Meeting of the Conference on Sciences and Techniques of Automatic Control and Computer Engineering (MI-STA)

PB - IEEE

T2 - 2022 IEEE 2nd International Maghreb Meeting of the Conference on Sciences and Techniques of Automatic Control and Computer Engineering (MI-STA)

Y2 - 23 May 2022 through 25 May 2022

ER -