• The Journal of Information Technology
• /
• v.4 no.1
• /
• pp.71-79
• /
• 2001

This paper presents an adaptive back propagation algorithm that update the learning parameter by the generated error, adaptively and configuration of the range for the initial connecting weight according to the different maximum target value from minimum target value. This algorithm is expected to escaping from the local minimum and make the best environment for the convergence. On the simulation tested this algorithm on three learning pattern. The first was 3-parity problem learning, the second was $7{\times}5$ dot alphabetic font learning and the third was handwritten primitive strokes learning. In three examples, the probability of becoming trapped in local minimum was reduce. Furthermore, in the alphabetic font and handwritten primitive strokes learning, the neural network enhanced to loaming efficient about 27%~57.2% for the standard back propagation(SBP).

• The Transactions of the Korea Information Processing Society
• /
• v.7 no.6
• /
• pp.1834-1843
• /
• 2000

In this paper, we propose a modular neural network based classification method for handwritten numerals utilizing rotated and translated images of an input image. The whole numeral pattern space is divided into smaller spaces which overlap each other and form multiple clusters. On these multiple clusters, multiple multilayer perceptrons (MLP) neural networks, specialized in those clusters, are constructed. Thus, each MLP acts as an expert network on the corresponding cluster. An MLP is also used as a gating network functioning as a mediator among the multiple MLPs. In the learning phase, an input numeral image is dithered by tow geometric operations of translation and rotation so that new numeral images similar to original one are generated. In the recognition phase, we utilize not only input numeral image, but also nearly generated images through the rotation and the translation of the original image. Thus, multiple output values for those generated images were combined to make class decision by various combination methods. The experimental results confirm the validity of the proposed method.

• The Journal of the Korea Contents Association
• /
• v.9 no.1
• /
• pp.159-166
• /
• 2009

In this paper, we analyse the noise robustness of MLPs(Multilayer perceptrons) through deriving the probability density function(p.d.f.) of output nodes with additive input noises and the misclassification ratio with the integral form of the p.d.f. functions. Also, we propose linear preprocessing methods to improve the noise robustness. As a preprocessing stage of MLPs, we consider ICA(independent component analysis) and PCA(principle component analysis). After analyzing the noise reduction effect using PCA or ICA in the viewpoints of SNR(Singal-to-Noise Ratio), we verify the preprocessing effects through the simulations of handwritten-digit recognition problems.

• Journal of KIISE:Computing Practices and Letters
• /
• v.16 no.5
• /
• pp.546-550
• /
• 2010

In this paper, we propose a method for the postprocessing of cursive script recognition in Wine Label Images. The proposed method mainly consists of three steps: combination matrix generation, character combination filtering, string matching. Firstly, the combination matrix generation step detects all possible combinations from a recognition result for each of the pieces. Secondly, the unnecessary information in the combination matrix is removed by comparing with bigram of word in the lexicon. Finally, string matching step decides the identity of result as a best matched word in the lexicon based on the levenshtein distance. An experimental result shows that the recognition accuracy is 85.8%.

• Journal of KIISE:Databases
• /
• v.36 no.5
• /
• pp.366-373
• /
• 2009

In this paper, we propose a similarity search algorithm for image databases. One of the central problems regarding content-based image retrieval (CBIR) is the semantic gap between the low-level features computed automatically from images and the human interpretation of image content. Many search algorithms used in CBIR have used the Minkowski metric (or $L_p$-norm) to measure similarity between image pairs. However those functions cannot adequately capture the aspects of the characteristics of the human visual system as well as the nonlinear relationships in contextual information. Our new search algorithm tackles this problem by employing new similarity measures and ranking strategies that reflect the nonlinearity of human perception and contextual information. Our search algorithm yields superior experimental results on a real handwritten digit image database and demonstrates its effectiveness.

• Journal of Institute of Control, Robotics and Systems
• /
• v.14 no.8
• /
• pp.818-823
• /
• 2008

In this paper, we propose a feature extraction method which extracts directional features of handwritten numerals by using the projection runlength. Our directional featrures are obtained from four directional images, each of which contains horizontal, vertical, right-diagonal and left-diagonal lines in entire numeral shape respectively. A conventional method which extracts directional features by using Kirsch masks generates edge-shaped double line directional images for four directions, whereas our method uses the projections and their runlengths for four directions to produces single line directional images for four directions. To obtain the directional projections for four directions from a numeral image, some preprocessing steps such as thinning and dilation are required, but the shapes of resultant directional lines are more similar to the numeral lines of input numerals. Four [$4{\times}4$] directional features of a numeral are obtained from four directional line images through a zoning method. By using a hybrid feature which is made by combining our feature with the conventional features of a mesh features, a kirsch directional feature and a concavity feature, higher recognition rates of the handwrittern numerals can be obtained. For recognition test with given features, we use a multi-layer perceptron neural network classifier which is trained with the back propagation algorithm. Through the experiments with the handwritten numeral database of Concordia University, we have achieved a recognition rate of 97.85%.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.41 no.6
• /
• pp.61-67
• /
• 2004

We propose a new algorithm to generate additional training patterns using the brain-style information processing algorithm, that is, supervised and unsupervised learning models. This will be useful in the case that we do not have enough number of training patterns because of limitation such as time consuming, economic problem, and so on. We adopt the independent component analysis as an unsupervised model for generating exempalr patterns and multilayer perceptions as supervised models for verifying usefulness of the generated patterns. After statistical analysis of the proposed pattern generation algorithm, we verify successful operations of our algorithm through simulation of handwritten digit recognition with various numbers of training patterns.

• Journal of the Korean Institute of Telematics and Electronics C
• /
• v.35C no.6
• /
• pp.67-75
• /
• 1998

Although the nCE(n-th order cross-entropy) error function resolves the incorrect saturation problem of conventional EBP(error back-propagation) algorithm, the performance of MLP's (multilayer perceptrons) trained using the nCE function depends heavily on the order of the nCE function. In this paper, we propose an adaptive learning rate to make the performance of MLP's insensitive to the order of the nCE error. Additionally, we propose a limited error signal of output node to prevent unstable learning due to the adaptive learning rate. The effectiveness of the proposed method is demonstrated in simulations of handwritten digit recognition and thyroid diagnosis tasks.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.11 no.5
• /
• pp.332-339
• /
• 1986

This paper proposes a new method for the normalization of shape pattern based on Gaussian probability density function to increase automatic recognition rate of hand-written Hangul pattern. The sizes of hand-written Hangul pattern are detected from the input images, and pattern sizes are normalized by two variables interpolation. The pattrn shapes are noralized by letting correlation coefficients equal to zero. It is analyzed theoretically and verified through computer simulation for the relation between input image and normaized shape pattern. It is confirmed that this method is effective and reasonable for deformed hand-written Hangul pattern. Experimental resu results show that the declination. size and stroke width of hand-written Hangul patterns are mych improved.

• The Journal of the Korea Contents Association
• /
• v.5 no.6
• /
• pp.57-64
• /
• 2005

LBL(Layer-By-Layer) algorithms have been proposed to accelerate the training speed of MLPs(Multilayer Perceptrons). In this LBL algorithms, each layer needs a error function for optimization. Especially, error function for hidden layer has a great effect to achieve good performance. In this sense, this paper proposes a new hidden layer error function for improving the performance of LBL algorithm for MLPs. The hidden layer error function is derived from the mean squared error of output layer. Effectiveness of the proposed error function was demonstrated for a handwritten digit recognition and an isolated-word recognition tasks and very fast learning convergence was obtained.