International Journal of Computer Applications (0975 – 8887) Volume 122 – No.12, July 2015 29 Neural Networks and Machine Learning for Pattern Recognition Arafat A. Muharram Associate Professor Department of Computer Science Faculty of Computer Science and Engineering Hodeidah University, Hodeidah, Y.R. Khaled M. G. Noaman Associate Professor Department of Distance Learning Deanship of E-Learning and Distance Learning, Jazan University Jazan, Kingdom of Saudi Arabia, K.S.A Ibrahim A.A. Alqubati Professor Department of Information Systems Community College Najran University Najran, Kingdom of Saudi Arabia, K.S.A ABSTRACT This paper represents an application study for using the Neural Networks and Machine Learning to recognize the English alphabet (A-Z) through the use of pattern recognition techniques in image processing and specifically to the application of Neural Networks and machine learning as a matrix two dimension. We used two techniques ANN and ML to compare their efficiencies and accuracies. We got 86.92% for ANN and 91.2% for ML. General Terms Pattern Recognition, Image Processing. Keywords Neural Networks, Machine Learning, Image Processing, Pattern Recognition. 1. INTRODUCTION Using the ANN to recognize images is of great importance to researchers in the ANN field. It is worth noting that the different shapes of images and their contents as well as dealing with the ANN to recognize images are so necessary for researchers that they have to process these images through the stages that these images have to pass through i.e. clearing the images (filtering ),minimizing or maximizing the images (scaling),the movement of the image at a specific angle (rotation),segmenting the images (segmentation),etc, depending on the nature of the image [1]. Since the applications on one of the ANNs i.e. back propagation which is used in this paper require data representation, that is meant for training in a particular way that helps in the ANN learning .It is necessary of course after the processing of the image used for training the ANN to represent the pixels value by the bipolar form (1,-1), which is one of the popular forms. After testing the ANN, we get the text file that contains the letters of the English alphabet[13 ]. In fact the text file that we get reduces the store size of the said letters. That is to say, each pixel consists of four bytes, whereas in the text file each letter consists of one byte only. It is also important to mention here that the structure of the ANN was designed keeping in mind the number of neurons and layers as well as the weights etc. Furthermore the program was written in a way that matches the designed structure of the ANN. The second technique we used was Machine Learning. The Machine Learning field evolved from the broad field of Artificial Intelligence, which aims to mimic intelligent abilities of humans by machines. An important task in Machine Learning is classification. Machine Learning in this paper was used to recognize the shape of the character ,which depend on the black pixels taken individually, especially those in the interior of the objects shown. They don't carry any information whether they are part of a representation of an object or another, it is their spatial interrelations which make us recognize the shape and to create variations such as fonts of different boldness of the object[2][7]. 2. 2. RESEARCH METHODOLOGY 2.1 Image Processing The techniques used in the image processing are as follows: 2.1.1 Segmentation This means dividing the image into small parts that are the constituents of the original images. Since this paper aims at the optical character recognition (OCR), if one image contains two characters (A, B), the segmentation of this image will result in two separated images (small parts) for (A) , (B) individually. 2.1.2 Filtering Filtering is of great importance as it vanishes the scattered spots as well as the noise found in the image. Thus we get a clean and pure image. In the present paper, I have used “Median” method to clear the image. This requires taking specific areas of the image with square dimension (2D), following up the values of all the pixels, arranging them, taking their median and replacing all the (pixels) by the median pixel [6] . 2.1.3 Image Minimization