• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.31B no.9
• /
• pp.159-165
• /
• 1994

In this paper, we propose a multi-layer associative neural network structure suitable for hardware implementaion with the function of performance refinement and improved robutst capability. Unlike other methods which reduce network complexity by putting restrictions on synaptic weithts, we are imposing a requirement of hidden layer neurons for the function. The proposed network has synaptic weights obtainted by Hebbian rule between adjacent layer's memory patterns such as Kosko's BAM. This network can be extended to arbitary multi-layer network trainable with Genetic algorithm for getting hidden layer memory patterns starting with initial random binary patterns. Learning is done to minimize newly defined network error. The newly defined error is composed of the errors at input, hidden, and output layers. After learning, we have bidirectional recall process for performance improvement of the network with one-shot recall. Experimental results carried out on pattern recognition problems demonstrate its performace according to the parameter which represets relative significance of the hidden layer error over the sum of input and output layer errors, show that the proposed model has much better performance than that of Kosko's bidirectional associative memory (BAM), and show the performance increment due to the bidirectionality in recall process.

• Korean Journal of Optics and Photonics
• /
• v.11 no.1
• /
• pp.47-57
• /
• 2000

A two-stage iterative Fourier transform algorithm, based on hybrid input-output algorithm and new Pnoise algorithm, is used to design continuous and quantized phase-only diffractive pattern elements which produce arbitrary given intensity patterns via Fraunhofer diffraction. Numerical results for two $128\times128$ binary patterns and two grayscale patterns are compared with those of other algorithms. It is found that the algorithm yields better signal-to-noise ratio and even better uniformity with slightly lower diffraction efficiency than other algorithms. We investigated the dependence of performance on parameters used in the algorithm, size of noise region, and the number of phase levels for quantized elements. In the case of quantized phase elements, the size of noise region plays a greater role in determining the performance of the algorithm than given intensity pattern itself. tself.

• Journal of Microbiology and Biotechnology
• /
• v.12 no.6
• /
• pp.909-915
• /
• 2002

The biodegradation of BTEX components (benzene, toluene, ethylbenzene, o-xylene, m-xylene, and p-xylene) individually and in mixtures was investigated using the o-xylene-degrading thermo-tolerant bacterium Ralsronia sp. strain PHS1 , which utilizes benzene, toluene, ethylbenzene, or o-xylene as its sole carbon source. The results showed that as a single substrate for growth, benzene was superior to both toluene and ethylbenzene. While growth inhibition was severe at higher o-xylene concentrations, no inhibition was observed (up to 100 mg $l^-1$) with ethylbenzene. In mixtures of BTEX compounds, the PHS1 culture was shown to degrade all six BTEX components and the degradation rates were in the order of benzene, toluene, o-xylene, ethylbenzene, and m- and p-xylene. m-Xylene and p-xylene were found to be co-metabolized by this microorganism in the presence of the growth-supporting BTEX compounds. In binary mixtures containing the growth substrates (benzene, toluene, ethylbenzene. and o-xylene), PHS1 degraded each BTEX compound faster when it was alone than when it was a component of a BTEX mixture, although the degree of inhibition varied according to the substrates in the mixtures. p-Xylene was shown to be the most potent inhibitor of BTEX biodegradation in binary mixtures. On the other hand, the degradation rates of the non-growth substrates (m-xylene and p-xylene) were significantly enhanced by the addition of growth substrates. The substrate utilization patterns between PHS1 and other microorganisms were also examined.

• Journal of KIISE
• /
• v.44 no.7
• /
• pp.692-702
• /
• 2017

Face detection using the LBP based feature descriptor has issues in that it can not represent spatial information between facial shape and facial components such as eyes, nose and mouth. To address these issues, in previous research, a facial image was divided into a number of square sub-regions. However, since the sub-regions are divided into different numbers and sizes, the division criteria of the sub-region suitable for the database used in the experiment is ambiguous, the dimension of the LBP histogram increases in proportion to the number of sub-regions and as the number of sub-regions increases, the sensitivity to facial orientation rotation increases significantly. In this paper, we present a novel facial region segmentation method that can solve in-plane rotation issues associated with LBP based feature descriptors and the number of dimensions of feature descriptors. As a result, the proposed method showed detection accuracy of 99.0278% from a single facial image rotated in orientation.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.37 no.5A
• /
• pp.375-387
• /
• 2012

This paper proposes a new real-time object tracking algorithm using particle filters with color and texture features in moving CCD camera images. If the user selects an initial object, this region is declared as a target particle and an initial state is modeled. Then, N particles are generated based on random distribution and CS-LBP (Centre Symmetric Local Binary Patterns) for texture model and weighted color distribution is modeled from each particle. For observation likelihoods estimation, Bhattacharyya distance between particles and their feature models are calculated and this observation likelihoods are used for weights of individual particles. After weights estimation, a new particle which has the maximum weight is selected and new particles are re-sampled using the maximum particle. For performance comparison, we tested a few combinations of features and particle filters. The proposed algorithm showed best object tracking performance when we used color and texture model simultaneously for likelihood estimation.

• KIPS Transactions on Software and Data Engineering
• /
• v.2 no.4
• /
• pp.273-280
• /
• 2013

This paper presents novel OCS-LBP (Oriented Center Symmetric Local Binary Patterns) based on orientation of pixel gradient and image retrieval system based on BoF (Bag-of-Feature) and random forest classifier. Feature vectors extracted from training data are clustered into code book and each feature is transformed new BoF feature using code book. BoF features are applied to random forest for training and random forest having N classes is constructed by combining several decision trees. For testing, the same OCS-LBP feature is extracted from a query image and BoF is applied to trained random forest classifier. In contrast to conventional retrieval system, query image selects similar K-nearest neighbor (K-NN) classes after random forest is performed. Then, Top K similar images are retrieved from database images that are only labeled K-NN classes. Compared with other retrieval algorithms, the proposed method shows both fast processing time and improved retrieval performance.

• Journal of the Korea Convergence Society
• /
• v.10 no.5
• /
• pp.27-34
• /
• 2019

In this paper, we propose a method to remove the background area by analyzing the pattern of the character area. In the character detection result of the MSER(Maximally Stable External Regions) method which distinguishes a region having a constant contrast background regions were detected. To solve this problem, we use the MSER method in natural images, the background is removed by calculating the change rate by searching the character area and the background area which are not different from the areas where the contrast values are different from each other. However, in the background removed image, using the LBP(Local Binary Patterns) method, the area with uniform values in the image was determined to be a character area and character detection was performed. Experiments were carried out with simple images with backgrounds, images with frontal characters, and images with slanted images. The proposed method has a high detection rate of 1.73% compared with the conventional MSER and MSER + LBP method.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.21 no.1
• /
• pp.269-280
• /
• 1996

We proposed a symbolic substitution method based on an optical correlator for an optical parallel addition. In the proposed symbolic substitution method, we used redundant binary number of the symbolic substitution rules as a number system and chose MAC3E filter which had very low sidelobes and good correlation peak compared with SDF filter as the optical correlation filter. We encoeded input numbers property to increase the discrimination capability and divided inpt patterns into 5 groups of the same addition results to minimize the number of symbolic substitution rules. Through the computer simulation, we confirmed the proposed method was suitable to implement the optical parallel adder.

• Journal of Information Technology Application
• /
• v.1
• /
• pp.173-211
• /
• 1999

The task of classification permeates all walks of life, from business and economics to science and public policy. In this context, nonlinear techniques from artificial intelligence have often proven to be more effective than the methods of classical statistics. The objective of knowledge discovery and data mining is to support decision making through the effective use of information. The automated approach to knowledge discovery is especially useful when dealing with large data sets or complex relationships. For many applications, automated software may find subtle patterns which escape the notice of manual analysis, or whose complexity exceeds the cognitive capabilities of humans. This paper explores the utility of a collaborative learning approach involving integrated models in the preprocessing and postprocessing stages. For instance, a genetic algorithm effects feature-weight optimization in a preprocessing module. Moreover, an inductive tree, artificial neural network (ANN), and k-nearest neighbor (kNN) techniques serve as postprocessing modules. More specifically, the postprocessors act as second0order classifiers which determine the best first-order classifier on a case-by-case basis. In addition to the second-order models, a voting scheme is investigated as a simple, but efficient, postprocessing model. The first-order models consist of statistical and machine learning models such as logistic regression (logit), multivariate discriminant analysis (MDA), ANN, and kNN. The genetic algorithm, inductive decision tree, and voting scheme act as kernel modules for collaborative learning. These ideas are explored against the background of a practical application relating to financial fraud management which exemplifies a binary classification problem.

• Journal of Biosystems Engineering
• /
• v.24 no.5
• /
• pp.431-438
• /
• 1999

The environmental control based on interactive thermoregulatory behavior for swine production has many advantages over the conventional temperature-based control methods. Therefore, this study was conducted to compare various feature selection methods using postural images of growing pigs under various environmental conditions. A color CCD camera was used to capture the behavioral images which were then modified to binary images. The binary images were processed by thresholding, edge detection, and thinning techniques to separate the pigs from their background. Following feature were used for the input patterns to the neural network ; \circled1 perimeter, \circled2 area, \circled3 Fourier coefficients (5$\times$5), \circled4 combination of (\circled1 ＋ \circled2), \circled5 combination of (\circled1 ＋ \circled3), \circled6 combination of (\circled2 ＋ \circled3), and \circled7 combination of (\circled1 ＋ \circled2 ＋ \circled3). Using the above each input pattern, the neural network could classify training images with the success rates of 96%, 96%, 96%, 100%, 100%, 96%, 100%, and testing images with those of 88%, 86%, 93%, 96%, 91%, 90%, 98%, respectively. Thus, the combination of perimeter, area and Fourier coefficients of the thinning images as neural network features gave the best performance (98%) in the behavioral classification.