The background motion presented in the image sequence normally consists of complex motion patterns and is affected by local moving objects. In this paper, a new method to estimate the affine modal parameters based on even sampled pixel blocks in frames is presented. It leads to great reduction in computational load. In order to estimate the background motion using only the background pixels and reduce the influence of moving objects to a great extend, Huber function is considered to discriminate between the background pixels and the foreground pixels. Experiment results show that it is a fast and robust method.

Pitch detection in noisy environment plays an important role in speech analyzing and recognition. Autocorrelation and average magnitude difference function are essential methods in pitch detection. A novel and effective method combining them is proposed. By simulation experiments, it is shown that the proposed pitch extraction method is feasible. It behaves more robustly than the conventional methods, especially effective at low signal－to－noise ratio.

An approach to face detection which based on skin color model and eigenfaces is introduced. With the stability of human skin color's distribution in the chromatic space, the skin color areas are segmented. Further investigation is made based on the SNR of their reconstruction image in the eigenface subspace. The result of the experiment proved to be promising.

Matlab has powerful abilities in matrix operation, equation solution, etc.. Based on Matlab, there is a large amount of simulations on varieties of control systems, but little simulation is on three dimension TV targets in a complicated background. Although many softwares simulating TV targets using virtual reality techniques have been developed, the predominance of Matlab on scientific research will be not fully exerted in this condition. If virtual reality techniques are used to simulate TV targets and Matlab is used to carry out research work, the two aspects can not be connected as an organic whole and the research work will be trivial and complicate, and the final software will also lack integrity. In this paper, using Matlab, 2D image of 3D TV target whose attitude angle in the sky can be described precisely can be given. It provides a practical method for simulating a 3D dynamic target under a complicated background in the sky. It is a good basis for extracting information of 3D target from 2D views.

Computer editing and generating of dynamic scenes of various targets and backgrounds have been actualized. At first, four elements composed the scenes are analyzed, i.e., target, background, moving pattern and special events. Then they are programmed into four modules to fulfil corresponding tasks. In the target and background modules, the images are processed and interest parameters are calculated. In the moving pattern module, the pattern is specified. In the events module, some typical events independent of target can be added. At last, using the synthesizing techniques, the required dynamic scenes are automatically generated by computer. The original target and background images can be easily got through the image DBMS.

The basic concept, main advantage, architecture, method and application of multi－sensor data fusion are described. Data fusion is the process of combining, correlating, and associating data from multi－sensor or other information sources to identity estimation and assessment of situations. Some general problems on fault diagnosis are introduced. The process of data fusion of fault diagnosis are also introduced. Their specialty and the layered fusion architecture fitting with fault diagnosis of control system are analyzed, and an architecture of fault diagnosis which is based on the information fusion are deeply investigated.

The design and principle of a self－adaptive fuse system are presented in this paper. Because the system can adapt to some uncertain fight conditions and parameters, it can precisely recognize target and realize the biggest destructivity of fight part with the best way, at the best time and the best distance. Some key problems in the system designing are analysed, including small field of view probing way of three branches, accurately taking of light signal criterion time, anti－jamming of ocean wave and recognizing the situation of ocean surface. The scheme diagram and software flow chart are also presented. All parameters of the system reach the designing requirement and are verified in the test.

Avalanched photodiode(APD) is useful in scanning laser imaging system due to their high sensitivity and fast responsibility. However, the optimal bias voltage of APD circuits is not fixed because of the changing environment conditions and temperature. It is proved that keeping APD at optimal operating state is an important condition for increasing detection sensitivity of optical receiver. Regarding this, a proper method by constant false alarm rate together with threshold detector and noise is provided. The voltages are adjusted to maintain the APD's optimal working status.

In this paper, the mechanism of intracavity optical parametric oscillator (OPO) is briefly analyzed. A high efficent and compact 1.57μm eye－safe intracavity optical parametric oscillator is developed, which is based on a non－critical phase－matched KTPⅡ crystal placed inside the resonator of a KD *P electronic－optical Q－switched (Ce, Nd):YAG laser by flashlamp pumping with water cooling. In experiment, the highest signal output 87.7mJ at 10Hz is obtained, the pulse width is about 7ns, the wall－plug efficiency is up to 5.8‰, and the OPO device is 410mm×130mm×80mm in size.

Zoom lens has increasing applications for its inherent properties. At present, the continuous zoomed lens generally adopt the mechanical－compensated optical configuration in which the given variable lenses groups move along the optical axis driven by a cam to achieve the continuous change of EFL , and it usually has two or more variable groups. Its cam optimization is the key to actualize the performance of its optical design and the zooming process, while the smoothness and speediness of zooming movement must be considered for a military use. The cam optimization method of a mechanical－compensated zoom lens with two variable groups is briefly described and the main body of the Macro－PLUS sequence which is suitable for different numbers of cam data points and optical design of the zoom lens with two variable groups are given.

The long－wave HgCdTe PC detectors were irradiated by laser beam. The power of the laser beam was changeable and below the permanent damnification threshold of the long－wave HgCdTe PC detectors. The resistance－temperature characteristic of the devices before and after the irradiation is measured, since the resistance－temperature characteristic of the devices can be used to study the electric parameters of the material. This method is used to fit the result of the experiment. It indicates that the proportion of the element Cd become larger. It is concluded that the electronic mobility and electronic concentration of the devices would be reduced after the break of the performance of the detectors due to the irradiation of laser beam.

As to paroxymal data in PCI bus, a method of the hardware and software design of MPEG－2 transport stream (TS) real－time decoder and TS sender are described in this paper. This method is realized by adding a buffer to the interface chip of PCI and based on a PCI card. The system sustains decoding compressed images in standard definition, more than one video outputting format. It's up to 40Mbps bit rate of outputting TS in the experiment.

The hardware interface and software design of MPEG－2 decoder card based on PCI local bus is presented. By analyzing program information of the multi－program MPEG－2 transport stream, the user could pick out one set of program to decode and display. The encoded data is sent to the decoder card through PCI local bus. The data exchange among memory, PCI bus, and hard disk were implemented by multi－thread processing in the software. The software design architecture is presented.

Compressed digital video data is usually packed to transport stream(TS) which has its own bit rate. In some situation, such as the front－end of channel modulating, it is required to adjust the bit rate of the TS. A hardware implemented, dual outputs, real－time and program selectable bit rate transforming system is introduced in this article.

The format of transport stream of high definition television(HDTV) is defined in MPEG－2 system standard ISO/IEC 13818－1.The function of timing information code on synchronizing the video and audio is explained in this paper,and the principle of synchronization of video and audio in HDTV decoder is analyzed.The problems in practical ASIC design, such as the mechanisms of rebuilding of system clock and repeating frame or skipping frame in video and audio, are discussed, a simple non－PLL digital scheme of synchronizing the video and audio is also provided. Update now the method of rebuilding system clock, lagged skipping frame method of synchronization of video and system clock, and digital PLL method of synchronization of audio and system clock are adopted in this scheme. The frame rate of video and warp of audio PCM clock frequency are discussed at last.

The general structure and the method of constructing index tree and typical query types in high－dimensional data space are investigated. Analytical comparison among several representative high dimensional indexing methods as K－D－Tree, R－Tree, R;+*－Tree, X－Tree, TV－Tree, Hilbert R－Tree is represented in such aspects as node shape, overlapping, insert principle, division principle and re－insert at the end.

The characteristics of a computerized radiography system were briefly discussed. The method for X－ray image digitalization was introduced. The hardware circuit for X－ray image data acquistion was given. Architecture of Windows NT and its kernal mode driver were briefly illustrated, and the working principle of I/O request in Windows NT was interpreted. The kernel mode driver for the X－ray image data acquisition card was designed with NT DDK. The real－time acquisition and storing of X－ray image data were carried out by the combination of hardware and software. In the long－term experiment, it has been proved that the method is feasible and reliable, and it can meet the practical commands.

IR stealth technology is reducing or changing the IR radiation characteristics of the object in order to achieve low detectionable characteristics of the object. It can reduce and absorb the IR radiation energy of the object by improving structure design and applying the principle of IR physics, thus it's difficult for IR detection devices to detect the object. The preliminary research and early development of IR stealth technology was accomplished in the late 1970's. It had important break through and entered the practical period from basic research period. Since 1980, IR stealth technology has been widely applied in the new weapons of foreign army, navy and airforce. At first, the principles of IR stealth are analyzed, then its application and developing tendency are studied.

In this paper, a method that uses the wavelet transform to suppress the speckle noise in synthetic aperture radar (SAR) images is described. Based on the previous wavelet approach, the properties of the wavelet coefficients have been taken into more account so that the edges information in images can be preserved well and the computational burden has been reduced at the same time. And this method is also compared with other classical denoising methods, such as median filter, Lee filter, Frost filter and so on. The results of the simulations and application to synthetic image and real SAR images prove the good performance of proposed new wavelet method.

The image texture includes both gray attributes and space relationship. In this paper, a new kind of weighted differential box－counting algorithm of fractal dimension is proposed. The value of weight is decided by gray scale. Experiments on meteorology nephogram from both GMS and NOAA satellite show that the algorithm is very efficient in automatic classification of cloud and mist. The correct rate of mist extraction of one image is above 90%.

Segmentation of handwritten Chinese characters is an important task for preprocessing of Chinese character recognition. A new algorithm for handwritten Chinese character segmentation based on extracting and knowledge－based merging of stroke bounding boxes is proposed in this paper. The knowledge－based merging operations are used to merge stroke bounding boxes according to the structure feature of Chinese characters. The experiment shows the new algorithm is very effective for off－line handwritten Chinese character segmentation.