Recently the need of person identification, which we human beings have performed with eyes and ears, using a computer is becoming larger and larger. It is a typical problem of pattern recognition and is very easy for humen beings, but it is not so easy for a computer. The fields that require person identification can be classified to three kinds. The first is credit business. the second is security business such as gate management. and the third is person collation using criminals database of the police. In this report, the recent trends and topics about person identification technique are shown.

View full abstract

In order to support many kinds of collaboration over long distances by using visual telecommunication, the transmission of visual information related to the participants are necessary. When people collaborate in the same workspace, they use visual interactions such as facial expressions and eye movements, as well as conversation. In order to realize coexistence in a collaborative workspace, support of such visual interactions is essential. Therefore, it is important to exactly take human area by video camera. This paper describes human area tracking methods in the collaborative workspace, and the tracking characteristics of human area by using real-time human area detection and tracking system.

View full abstract

This paper deals with a segmentation method of an image composed of some kinds of textures with randomness by using neural network. After a considering image is divided into many small regions with the same size, the feature vectors of small regions are extracted by fitting two dimensional autoregressive models to the data in each regions. The clustering of feature vectors is performed by applying self organization algorithm developed by Kohonen, and the result of clustering is mapped to the original image. Furthermore a method which applies backpropagation algorithm to the result of clustering is proposed in order to improve the accuracy of segmentation.

View full abstract

Multiplex transmission using PPM and PAM is considered. Information included in pulse amplitude and position can be separated by the optimum linear filter for PPM transmission. From this viewpoint, Multiplex transmission systems that use PPM and PAM at the same time can be realized. First, we show the optimum pulse timing detector and the structure of PPM-PAM system. Following, PPM-PAM demodulator is considered. we also consider the filter, system synchronization, analog/digital transmission and adoption of this multiplex transmission in short.

View full abstract

In this paper, we describe new noncoherent FSK receiver by variable threshold. It is composed of a second-order band pass filter (BPF) of butterworth type, a first-order digital PLL demodulator and a decision circuits. In general, we can make the following statement. If the bandwidth of above BPF becomes narrow, the SNR is improved under the condition of small Inter Symbol Interference (ISI), and is improved, and has maximum value, then SNR begin to decrease because of large ISI. Now, We explain the method which decreases ISI by making variable threshold level correspond to previous demodulated data. We draw the conclusion that The SNR is improved about 2.5 dB by using variable threshold level on the result of analys is of numerical calculation. Also as compared with GMSK, our receiver has approximately same ability. As a result of experiment, it obtained agreed approximately with those numerical calculation.

View full abstract

New FSK demodulation for fast frequrncy-hopped spread spectrum communication is proposed. We can not expect high hopping-rate of a conventional type using narrow band hand-pass filter and a envelope detector. Becase, by the FSk signals added a frequency bust of high frequency, a long tranjent phenomenon is induced. On the other hand. new demodulator is comprised of a band-pass filter with broad band, a PLL and integrate-dump filter using hopping-locked signal. Good performances are obtained by controled initial phase of the PLL.

View full abstract

Adaptive KL transform coding, which selects the best matched matrix pair from a prepared set of KL transform matrcies, achieves a higher S/N than that achieved by the DCT. The adaptive KL transform coding with variable maricies can trace the local characteristics of the image that is not stationary. It can be optimally designed on the minimum meam squre error criterion. KL transform coding methods are reviewed from the view point of optimality and the adaptive KL transform coding is categorized in the context. The difference between KL transform coding and SVD coding which also uses eigen vectors is considered

View full abstract

Until now, when we used the fuzzy c-means (FCM) algorithm, we had used the Euclidean norm as a measure of distance or dissimilarity. While Gustafson and Kessel suggested the fuzzy covariance (FC) clustering algorithm, which uses the Maharanobis norm as a measure of distace. This algorithm uses the fuzzy covariance matrix in the norm-inducing matrix. So it is useful for the cluster shape discriptions. In addition to the cluster shape, we suggest a new clustering method by considering the volume of clusters. Using this method, we can get more appropriate result.

View full abstract

This paper describes a high-resolution magnetoencephalographic mapping method using the multichannel SQUID system and the MRI. The brain magnetic field is considered to be produced by current dipoles. The localization of current sources is performed by computing using Biot-Savart's equation based on MEG. After corresponding the coordinate of SQUID system to the coordinate of MRI, each current dipole location is mapped into the anatomical MRI data.

View full abstract

A 3-D perception of the brain structures is useful for accurate quantitative diagnosis, for surgical or treatment planning. This paper presents a 3-D display method of medical images. This organ extraction method consists of a combination of multichannel MR images, the fuzzy c-means clustering and the object-connectivity check. This shading uses a volume rendering. However, better perception method than the 3-D display is desired for complicated organ shapes. A 3-D brain structure model is fabricated using MRI and laser lithography system. A complicated discrimination of gray matter, ventricle, basal ganglia, tumor, blood vessels, and their 3-D spatial relationships can be shown.

View full abstract

This paper describes a 3-D reconstruction method of cerebral blood vessels using a high speed X-ray rotational projection system. The measurement system uses a cone beam X-ray source and an image intensifier (II). The 3-D reconstruction method consists of correction of image distortion, estimation of geometrical parameters for measurement system, 3-D reconstruction from projections, and visualization of 3-D blood vessels. The proposed method is applied to 2-D projection images of cerebral blood vessels and its effectiveness is demonstrated.

View full abstract