Hybridized Convolution Neural Network and Multiclass-SVM Model for Writer Identification

Writer identification is a quite interesting research problem in the field of writing recognition due to an ambiguous written styles of different writers. This paper proposes a model that hybridises Convolutional Neural Network (CNN) and Multiclass-Support Vector Machine (MSVM) for getting a better accuracy in writer identification using English/Arabic handwriting samples. Deep identifying writer takes local handwritten image as input and CNN used for feature extraction then classified using MSVM classifier based on the extracted features from the CNN layers. The used CNN architecture was applied with multiple kernel sizes and each time the corresponding processing time and the identification accuracy was measured. The proposed system was applied over two publicly databases Khatt as an Arabic database and IAM as an English database and able to achieve an accuracy of around 99.8% for a set of 206 writers. The performance of the proposed system was compared with other existing writer identification systems.