Optical character recognition (OCR) is a study of how people use the images of the printed or handwritten words to enable computers to achieve the learning, recognition and decision. In the field of OCR, handwritten digit recognition is a technology about how people exploit computers to automatically identify handwritten Arabic number, which has developed rapidly in recent years. Handwritten digit recognition has high theoretical value and practical value in the field of information processing, machine translation and artificial intelligence. Therefore, it has gained much attention of researchers in science and technology fields. Researches on handwritten digit recognition have vast practical applications and financial implications [1]. At present, the recognition of handwritten digits is generally regarded as a branch of pattern recognition and realized mainly using the methods of Artificial Neural Network, Support Vector Machine and K-Nearest-Neighbors.
Pattern recognition refers to the process of processing and analyzing the various forms (numerical, literal, and logical) information that characterize things or phenomena and the process of describing, identifying, classifying, and interpreting things or phenomena, which is an important part in information science and artificial intelligence. Pattern recognition includes two parts. The first part consists of the modules of data acquisition, preprocessing, feature extraction and classification decision, realizing the classification of unknown patterns. The second part includes the module of classifier design, which is used to determine the parameters of classifier during the training process [1]. In the field of pattern recognition, there are some hot issues, such as character recognition, pedestrian detection and face recognition, are essentially the problems of image recognition. The system of image recognition usually utilizes the optical equipment and computers to analyze the image. On the basis of pattern recognition, the procedures of image recognition consist of image preprocessing, image segmentation, image feature extraction and image classification.
As a new pattern recognition technology, artificial neural network (ANN) has been applied more and more widely in recognition problems. ANN is a computational model based on a large collection of artificial neurons used in computer science and other research disciplines. A neural network is characterized by 1) its pattern of connections between the neurons, 2) its learning algorithm of determining the weights on the connections, and 3) its activation function [2]. In recent years, ANN has been a hot research topic in many disciplines. A great number of studies in these fields demonstrate the superiority of ANNs in terms of pattern recognition, signal processing, automatic control, computer vision and so on.
There are some methods of machine learning and feature extraction to solve the problem of the recognition of handwritten digits. Shahrezea et al. (1995) [3] used the shadow coding method for recognition of Persian handwritten digits. In the system proposed by Hosseini and Bouzerdoum (1996) [4], the digit images are represented by 11 line segments. The final features are extracted by calculating the quantitative values corresponding to each of these lines and then combining these values in some specific ways. In the paper written by Said et al. (1999) [5], the digit image size is first normalized into 16 × 20. Then the pixel values of the normalized image are fed into a neural network for classification. In the other paper, presented by Sadri et al. (2003), the first and last profiles of the image at both vertical and horizontal orientations are obtained in feature extraction stage. Then each of these profiles is represented as a one-dimensional signal. The derivative of each of these signals is represented by a vector of size 16, where these vectors constitute the feature vector.
The method proposed by Shahrezea et al. (1995) reached the recognition rate of 97.8%, while the recognition rates achieved by the others Hosseini and Bouzerdoum (1996), Said et al. (1999), Sadri et al. (2003) were less than 95%. Therefore, the accuracy of handwritten digit recognition remains low, and methods are typically customized for a particular script. Furthermore, the digital stroke is simple and the difference between strokes is relatively small, thus it is rather difficult to distinguish between certain numbers accurately. These methods cannot effectively solve this problem. In addition, there are often deformed characters, which cannot be recognized with above methods. Therefore, it is pertinent to propose a new method to solve the problem of classifying handwritten digits.
The primary objective of this paper is to make computers recognize the handwritten digits automatically by an effective method. To realize this goal, a neural network algorithm was proposed and tested on a database consisting of collected pictures of different handwritten digits. And then the comparison between the proposed method and the logistic regression method is done by analyzing data related to recognition accuracy and speed. Finally, the neural network method proves to be more effective and efficient for handwritten digits recognition.