Seongwon Cho

Iris Recognition Using Wavelet Features

The traditional iris recognition systems require equal high quality human iris images. A cheap image acquisition system has difficulty in capturing equal high quality iris images. This paper describes a new feature representation method for iris recognition robust to noises. The disc-shaped iris image is first convolved with a low pass filter along the radial direction. Then, the radially smoothed iris image is decomposed in the angular direction using a one-dimensional continuous wavelet transform. Each decomposed one-dimensional waveform is approximated by an optimal piecewise linear curve connecting a small set of node points. The set of node points is used as a feature vector. The optimal approximation procedure reduces the feature vector size while maintaining recognition accuracy. The similarity between two iris images is measured by the normalized cross-correlation coefficients between optimal curves. The similarity between two iris images is estimated using mid-frequency bands. The rotation of one-dimensional signals due to the head tilt is estimated using the lowest frequency component. Experimentally we show the proposed method produces superb performance in iris recognition.

ENSEMBLE COMPETITIVE LEARNING NEURAL NETWORKS WITH REDUCED INPUT DIMENSION

Conventional neural networks utilize all the dimensions of the original input patterns for training and classification. However, a particular attribute of the input patterns does not necessarily contribute to classification and may even cause misclassification in certain cases. A new ensemble competitive learning method using the reduced input dimension is proposed. In contrast to the previous ensemble neural networks which adjust learning parameters, the proposed method takes advantage of the information in each dimension of the input patterns. Since the degree of contribution of each attribute to classification is not known beforehand, the different input data sets with one dimension reduced are presented to multiple neural networks. The classification information from each competitive learning neural network is then combined to make a final decision for classification. In order to improve classification accuracy, the ambiguous output neurons are eliminated which cannot be assigned to any class after training. We use three consensus schemes to judge the classification using ensemble neural networks. The experimental results with remote sensing and speech data indicate the improved performance of the proposed method.

Iris recognition using LVQ neural network

In this paper, we discuss human iris recognition, which is based on iris localization, feature extraction, and classification. The features for iris recognition are extracted from the segmented iris pattern using two-dimensional (2-D) wavelet transform based on Haar wavelet. We present an efficient initialization method of the weight vectors and a new method to determine the winner in LVQ neural network. The proposed methods have more accuracy than the conventional techniques.

Self-organizing map with time-invariant learning rate and its exponential stability analysis

Abstract In this paper a new self-organizing map (SOM) is developed that is suitable for digital hardware implementation. In the proposed neural model, a time-invariant learning rate is used, whereas the original Kohonen SOM uses a time-varying learning rate. There is a binary re-inforcement term in order to compensate for the lowered learning ability due to the constant learning rate. The proposed SOM is exponentially stable. The experimental results conducted with two different types of data show that the proposed method has better learning ability than the original SOM.

Fuzzy Power Factor Control Systems

A method for obtaining the power energy with high quality is to keep the power factor for a load as close to unity as feasible. In this paper, we present a new method to improve the power factor for a load. The proposed method uses fuzzy control techniques in order to determine how many parallel capacitors are to be connected to the load for the correction of the power factor.

Multiple neural networks using the reduced input dimension

An ensemble of neural networks with competitive learning and consensus schemes is proposed. Conventional learning methods utilize all the dimensions of the original input patterns. However, a particular attribute of the input patterns does not necessarily contribute to classification. In this paper, we use the reduced input dimension for learning a neural network. We have developed three consensus schemes so as to judge the classification using multiple neural networks. The experimental results with remote sensing data indicate the improved performance of the networks when applying the proposed method to the conventional competitive learning algorithms.<<ETX>>

Iris recognition using a low level of details

This paper describes a new iris recognition algorithm, which uses a low level of details. Combining statistical classification and elastic boundary fitting, the iris is first localized. Then, the localized iris image is down-sampled by a factor of m, and filtered by a modified Laplacian kernel. Since the output of the Laplacian operator is sensitive to a small shift of the full-resolution iris image, the outputs of the Laplacian operator are computed for all space-shifts. The quantized output with maximum entropy is selected as the final feature representation. Experimentally we showed that the proposed method produces superb performance in iris segmentation and recognition. Index Terms: iris segmentation, iris recognition, shift-invariant, multiscale Laplacian kernel.

Eye Localization based on Multi-Scale Gabor Feature Vector Model

Eye localization is necessary for face recognition and related application areas. Most of eye localization algorithms reported thus far still need to be improved about precision and computational time for successful applications. In this paper, we propose an improved eye localization method based on multi-scale Gator feature vector models. The proposed method first tries to locate eyes in the downscaled face image by utilizing Gabor Jet similarity between Gabor feature vector at an initial eye coordinates and the eye model bunch of the corresponding scale. The proposed method finally locates eyes in the original input face image after it processes in the same way recursively in each scaled face image by using the eye coordinates localized in the downscaled image as initial eye coordinates. Experiments verify that our proposed method improves the precision rate without causing much computational overhead compared with other eye localization methods reported in the previous researches.

Motion detection in complex and dynamic backgrounds

For the detection of moving objects, background subtraction methods are widely used. In case the background changes, we need to update the background in real-time for the reliable detection of foreground objects. An adaptive Gaussian mixture model (GMM) combined with probabilistic learning is one of the most popular methods for the real-time update of the complex and dynamic background. However, the probabilistic learning approach does not work well in high traffic regions. In this paper, we classify each pixel into four different types: still background, dynamic background, moving object, and still object, and update the background model based on the classification. For the classification, we analyze a sequence of frame differences at each pixel and its neighborhood. We experimentally show that the proposed method learn complex and dynamic backgrounds in high traffic regions more reliably, compared with traditional methods.

Real-time audio surveillance system for PTZ camera

In this paper, we propose an audio surveillance system to detect and localize dangerous sound in real-time so as to be able to direct a PTZ camera to catch a snapshot image about the location of sound source instantly. The proposed audio surveillance system firstly detects foreground sound based on adaptive Gaussian mixture background sound model, and classifies it into one of pre-trained classes of foreground sounds based on GMM model. Next, it decides whether it belongs to dangerous class group or not. If it does, then a sound source localization algorithm based on Dual delay-line algorithm is applied to localize the sound source. Finally, the proposed system uses the sound source location information to pan and tilt the PTZ camera towards the orientation of the dangerous sound source, and take a snapshot against over the sound source region. Experiment results show that the proposed system can detect foreground sound stably and recognize dangerous sounds with a precision of 79% while the sound source localization can estimate orientation of the sound source with acceptably small error.