Comparison of different neural network architectures for digit image recognition (original) (raw)
Related papers
Identification of handwritten digits is one of the major areas of research in the field of character recognition. Artificial Neural Networks helps in computer vision that deals with how a computer could achieve high-level of understanding from digital images or videos. Thus, neural networks prove to be a boon in recognizing handwritten digits that are scanned as images. However, this paper aims at studying the working of specifically three neural networks-Multi-Layered Perceptron (MLP), Radial Basis Function (RBF) and Convolutional Neural Network (CNN). In order to focus majorly on the implementation of these three neural networks rather than the complexity of the dataset being used, we have used MNIST (Modified National Institute of Standard and Technology) dataset from keras library. The MNIST dataset contains 70,000 black and white images of handwritten English digits (60,000 training images and 10,000 testing images). In our study of the above three mentioned neural networks, we have used relu as activation function in the hidden layers and softmax as activation function in the final layer of neural network, Adam as an optimizer and cross-entropy as loss function. We have observed that all three networks give accuracy above 95%, however, the major difference is in its training time and error rate.
Performance Evaluation of Multi-Layer Perceptron (MLP) and Radial Basis Function (RBF
International Conference of the IEEE Nigeria Computer Chapter (NigeriaComputConf), 2019
The importance of character recognition cannot be over emphasized. It finds applications in many automated system. In most cases these applications require high precision (e.g. automatic grading system, document digitization, license plate recognition systems, e.t.c) as well as low resource overhead. However, these are conflicting requirements, because the more the precision required, the more computation needed hence the more increase in resource overhead. In the research, two classification algorithms in Artificial Neural Networks (ANN): Multi-Layer Perceptron (MLP) and Radial Basis Function (RBF) were applied to handwritten digit recognition and their performance is investigated. The duo were compared in terms of resources requirement for training and accuracy. It is found that MLP-NN is much faster to train (5.5min) compared to RBF (50.0min). However, during testing it is found that both have an accuracy of ≈ 95%.
2000
This chapter presents an axiomatic approach for reformulating radial basis function (RBF) neural networks. With this approach the construction of admissible RBF models is reduced to the selection of generator functions that satisfy certain properties. The selection of specific generator functions is based on criteria which relate to their behavior when the training of reformulated RBF networks is performed by gradient descent. This chapter also presents batch and sequential learning algorithms developed for reformulated RBF networks using gradient descent. These algorithms are used to train reformulated RBF networks to recognize handwritten digits from the NIST databases.
Neural Network Based Handwritten Digits Recognition- An Experiment and Analysis
International Journal of Computer and Electrical Engineering, 2009
Handwritten digit recognition has become very useful in endeavors of human/computer interaction. Reliable, fast, and flexible recognition methodologies have elevated the utility. This paper presents an experiment and analysis of the Neural Network classifier to recognize handwritten digits based on a standard database. The experimental setup implemented in Matlab determines the ability of a Multi-Layer Neural Network to identify handwritten digit samples 5-9. This network is the representative for recognition of remaining digits 0-4. We consider not only accurate recognition rate, but also training time, recognition time as well as the complexity of the networks. The Multi-Layer Perceptron Network (MLPN) was trained by back propagation algorithm. Network structures vary with the hidden units, learning rates, the number of iterations that seem necessary for the network to converge. Different network structures and their corresponding recognition rates are compared in this paper to find the optimal parameters of the Neural Network for this application. Using the optimal parameters, the network performs with an overall recognition rate 94%.
Handwritten Digits Recognition Using a Multilayer Feed-Forward Backpropagation Neural Network
Nowadays, many researchers are trying to build a program that can recognize handwritten digits, so that can be used in many various field. For example, it can be applied to read checks in banks or numbers in car plates but it is a challenging problem that need an accurate classification program. Our mission is to let the computer receive and isolate the digits individually and then interpret every digit by using Artificial Neural Network(ANN) based on feedforward back propagation algorithm. Therefore, this approach needs a very precise data in order to train the artificial neural network to Increase the accuracy of neural network classification. This paper presents an approach of recognizing handwritten digit by using three stages which are image processing, neural network recognition which is the major mission, and displaying the result. We create a system for dealing with such challenging problem. The system started by acquiring an image containing digits, this image was digitized and modified, so it can be sent to the ANN in order to be recognized by using feed forward back propagation algorithm. The studies were conducted on the handwriting digits of 80 independent writers who contributed a total of 1593 isolated digits these digits divided into two data sets: Training 1115 digits, testing 478 digits. An overall accuracy meet using this system was 92% on the test data set used.
Digit recognition using decimal coding and artificial neural network
Current artificial neural network image recognition techniques use all the pixels of an image as input. In this paper, we present an efficient method for handwritten digit recognition that involves extracting the characteristics of a digit image by coding each row of the image as a decimal value, i.e., by transforming the binary representation into a decimal value. This method is called the decimal coding of rows. The set of decimal values calculated from the initial image is arranged as a vector and normalized; these values represent the inputs to the artificial neural network. The approach proposed in this work uses a multilayer perceptron neural network for the classification, recognition, and prediction of handwritten digits from 0 to 9. In this study, a dataset of 1797 samples were obtained from a digit database imported from the Scikit-learn library. Backpropagation was used as a learning algorithm to train the multilayer perceptron neural network. The results show that the proposed approach achieves better performance than two other schemes in terms of recognition accuracy and execution time.
Simplified Neural Network Design for Hand Written Digit Recognition
Neural Network is abstraction of the central nervous system and works as parallel processing system. Optimization, image processing, Diagnosis and many other applications are made very simple through neural networks, which are difficult and time consuming when conventional methods are used for their implementation. Neural Network is the simplified version of human brain. Like human brain, neural networks also exhibit efficient performance on perceptive tasks like recognition of visual images of objects and handwritten characters etc: Recognition of handwritten digits is one of the oldest applications of ANN. The recognition of digits written in different handwritings and also from scanned text has remained a trouble thus it has received much attention of researchers in the field of artificial neural networks. We can distinguish among handwriting of different persons due to the fact that human brain is capable to even slight variations of visual images. In this research work a very simple and flexible neural network scheme is proposed and implemented for handwritten digit recognition, which will assist beginners and A.I students who want to understand perceptive capability of neural network. In the proposed system, a very simple design of artificial neural networks is implemented. First of all learning mechanism of the neural network is described and then its architecture is discussed. Proposed network is trained in supervised manner using various (approx: 250) patterns /fonts of handwritten digits. Unique token is allocated to digit when it is made input to the system. Network becomes adaptive when different patterns of the same digit are taught to the network for one particular token.
Designing of Digits Recognition Technique Using Neural Network
Handwriting Digit Recognition (HDR) has been research widely and there are many associated work in pattern recognition. Different commercial software's are also available in the Market. The major focus is on the improvements in the accuracy levels. In this paper, we have investigated different HDR systems and their implementation for digits in English. Lastly we have experimented digit recognition through artificial neural network using supervised deep learning.
Artificial Neural Network Classification for Handwritten Digits Recognition
Handwritten recognition is very powerful technology to support many applications comes in the forefront of automated sorting of letters and bank checks, and help the blind and Who have difficulty to read books and magazines, and the translation of books from one language to another, and converted to texts can store and processed in the computer. This paper is present two artificial neural network classification for handwritten digit recognition (from 0 to 9) with accuracy more than 98% by using an application of feed-forward multilayer neural network with two different classifiers (Forward Multilayer Neural Network FMNN and Binary Coding Neural Network BCNN).The highest recognition reliability and minimal error rate for the recognition of handwritten digits have been achieved. The back propagation algorithm minimizing the total error of the network over a set of training by searching of the weight value that achieves the objective. Binary coding approach is used to reducing the numb...
A Comparison of Architectural Varieties in Radial Basis Function Neural Networks
Representation of knowledge within a neural model is an active field of research involved with the development of alternative structures, training algorithms, learning modes and applications. Radial Basis Function Neural Networks (RBFNNs) constitute an important part of the neural networks research as the operating principle is to discover and exploit similarities between an input vector and a feature vector. In this paper, we consider nine architectures comparatively in terms of learning performances. Levenberg- Marquardt (LM) technique is coded for every individual configuration and it is seen that the model with a linear part augmentation performs better in terms of the final least mean squared error level in almost all experiments. Furthermore, according to the results, this model hardly gets trapped to the local minima. Overall, this paper presents clear and concise figures of comparison among 9 architectures and this constitutes its major contribution.