2019 International Conference on Computing, Mathematics and Engineering Technologies (iCoMET 2019) Size Invariant Handwritten Character Recognition using Single Layer Feedforward Backpropagation Neural Networks Adeel Yousaf 1 , Muhammad Junaid Khan 3 , Muhammad Jaleed Khan 2 , Nizwa Javed 2 , Haroon Ibrahim 2 , Khurram Khurshid 2 , Khawar Khurshid 4 1 Department of Aeronautics & Astronautics, Institute of Space Technology, Islamabad, Pakistan. 2 Military College of Signals, National University of Science and Technology, Rawalpindi, Pakistan. 3 WiSP Lab, Department of Electrical Engineering, Institute of Space Technology, Islamabad, Pakistan. 4 School of Electrical Engineering & Computer Science, National University of Science and Technology, Islamabad, Pakistan. adeel.yousaf@ist.edu.pk, mjk0853@gmail.com, mjk093@gmail.com, niz.jvd@gmail.com, haroon.ibrahim@ist.edu.pk, khurram.khurshid@ist.edu.pk, khawar.khurshid@seecs.edu.pk Abstract- Handwritten character recognition is among the most challenging research areas in pattern recognition and image processing. With everything going digital, applications of handwritten character recognition are emerging in offices, educational institutes, healthcare units and banks etc., where the documents that are handwritten are dealt more frequently. In this paper, a recognition system based on neural network that follows offline handwritten characters has been proposed for Latin digits and alphabets. Each of the characters that are extracted through query image is then resized dynamically to 60x40 pixels’ size and is then passed to the neural networks for the process of recognition. Dynamic resizing enables size invariance in the proposed system and also maintains the aspect ratio of the character so that the image is not distorted during resizing. Neural networks are trained with 19,422 English alphabets’ sample and 7,720 digits’ sample that are written through 150 different writers in various styles of handwriting. Experimental study realized very encouraging results which are compared with the modern methods on this subject corridor. Keywords- Handwritten Digit Recognition, Dynamic Resizing, Neural Networks, Handwritten Character Recognition, Hand-filled form processing. I. INTRODUCTION Handwritten Character Recognition (HCR) has been the most complicated and intriguing research area in the areas of digital image processing and pattern recognition [1-2]. An accurate system of recognition makes major contribution in making improvement in intelligent and automatic systems to enhance the interface between humans and technology in various applications. The major focus of the modern research is to develop algorithms that minimize the processing time while maintaining high accuracy rates [3]. Broadly, HCR systems are divided into two categories: online and offline. The online recognition systems are dependent on two of the coordinates expressed as the function of number and time and also the strokes’ order that writer made. Special hardware like a tablet PC and a pen is required in case. Experiments have proven that some of the online procedures are more accurate when referring to character recognition when compared with other offline methods because of additional time dependent information that is seen in the other method [4-5]. In offline recognition, the handwriting is accessible as an image. Offline recognition is attained through different applications like processing of bank cheque, handwritten form processing and document reading etc. That is why, offline systems are the major areas of research that have been used for making improvement in rates of recognition [6- 7]. Thus more research is being carried out to design a robust system for HCR and various methodologies are being explored to find out one that gives desired results for some particular application. For example, Artificial Neural Networks have now become more effective in making improvement in the accuracy of character recognition in offline systems because of accuracy, speed and robustness [8]. The fundamental steps of any HCR system include extraction of features, segmentation, pre-processing and matching. Pre- processing the image improves segmentation where individual handwritten characters are extracted from the text. The system accuracy is seen more dependent over the choice of the appropriate method of feature extraction. Different methods of feature extraction have been given in the literature like gradient features, projection histograms, template matching, Fourier descriptors, Gabor features, Zernike moments, unitary image transforms, geometric moment invariants, approximation of spline curve, description of graph and deformable templates [9]. Some of the important methods of classification that are used for HCR are kernel methods, statistical procedures dependent on Bayesian decision theory, ANN that comprise Support Vector Machines (SVMs), Hidden Markov Models (HMMs) and other different kinds of classifier combinations [10-11]. This paper expresses one layer feed forward neural network back-propagation dependent on the algorithm of HCR that employs a linked component dependent aspect for handwritten text’ segmentation from images of document. First of all, the pre- processing of query image is done for making it more compatible with the present system. Through image, Connected Components (CCs) are extracted for segmenting out the candidates of handwritten character. After this, some of the geometric checks are performed for filtering out such objects that are non-text through the detected CCs’ set. After the process of filtering, the CCs that tend to be the characters that are handwritten get dynamically resized to 60x40pixels. Dynamic resizing is employed to maintain the aspect ratio of the CC so that the image is not distorted during resizing. The resized CCs are passed to a well-trained neural network for recognition. Moreover, we have focused on various architectures and parameters of neural networks and have performed comparisons accordingly. The arrangement of rest of the paper has been done as follows. Section 2 refers to the related work. Section 3 refers to the proposed design in more detail. In section 4, the experimentation