Motion estimation is the key issue in video compressing. Several methods for motion estimation based on the center biased strategy and minimum mean square error trend searching have been proposed, such as TSS, FSS, UCBDS and MIBAS, but these methods yield poor estimates or find local minima. Many other methods predict the starting point for the estimation, these can be fast but are inaccurate. This study addresses the causes of wrong estimates, local minima and incorrect predictions in the prior estimation methods. The Multiple Searching Trend (MST) is proposed to overcome the problems of ineffective searches and local minima, and the Adaptive Dilated Searching Field (ADSF) is described to prevent prediction from wrong location. Applying MST and ADSF to the listed estimating methods, such as UCBDS, a fast and accurate can be reached. For this this reason, the method is called CockTail Searching (CTS). In another proposed method, we try to define the new criteria used to determine a referent macro block within the search window in a referent frame, which matches the estimated current macro block in the current frame, in motion estimation process used in MPEG standard. The Prediction Error(PE) in the Pixel Difference(PD) between the referent macro block and the current macro block is defined to be a new criterion which can get better performance in compressed data length than the Mean Square Error(MSE) used by most of motion estimation methods. The other criterion combined PE and MSE is proposed to get better performance than the PE. Two new criteria is applied to a famous motion estimation method, UCBDS, to show the performance of the new criteria. The evaluation results show that using new criteria in UCBDS can get more 40% reduction in compressed data size than the UCBDS with MSE.

This paper is concerned with the problem of finding an experimental design for discrimination between two rival models and for model robustness that minimizing the maximum bias simultaneously in binary response experiments. The criterion for model discrimination is based on the $T$-optimality criterion proposed in Atkinson and Fedorov (1975), which maximizes the sum of squares of deviations between the two rival models while the criterion for model robustness is based on minimizing the maximum probability bias of the two rival models. In this paper we obtain the optimum designs satisfy the above two criteria for some commonly used rival models in binary response experiments such as the probit and logit models etc.

¡@¡@The purpose of this research is to employ the STAR model in discussing and analyzing the relationship between stock index and macroeconomic variables in Taiwan, Japan and Korea. ¡@¡@Monthly stock market index data is analyzed over the period January 1990 to December 2000, with the sample period from January 2001 to April 2005 being used in an out-of -sample forecasting exercise. The macroeconomic variables considered in this paper include money supply, consumer price index, industrial production index, interest rate and exchange rate. ¡@¡@The empirical results of Taiwan, Japan and Korea show that LSTAR & ESTAR model improve both the in-sample fit and out-of-sample forecast of the data over both the linear model alternative.

This study includes three issues about the dynamic of 30-days Taiwan Commercial Paper rate (CP2).The first issue focuses on the estimation of continuous-time short-term interest rate models. We discretize the continuous-time models by using two different approaches, and then use weekly and monthly data to estimate the parameters. The models are evaluated by data fit. We find that the estimated parameters are similar for different discretization approaches and would be more stable and efficient under quasi-maximum likelihood (QML) with weekly data. There exists mean reversion for Taiwan CP rate and the relationship between the volatility and the level of interest rates are less than 1 and smaller than that of American T-Bill rates reported by CKLS (1992) and Nowman (1997). We also find that CIR-SR model performs best for Taiwan CP rate. The second issue compares the continuous-time short-term interest rate models empirically both by predictive accuracy test and encompassing test. Having the estimated parameters of the models by discretization of Nowman(1997) and QML, we produce the forecasts on conditional mean and volatility for the interest rate over multiple-step-ahead horizons. The results indicate that the sophisticated models outperform the simpler models in the in-sample data fit, but have a distinct performance in the out-of-sample forecasting. The models equipped with mean reversion can produce better forecasts on conditional means during some period, and the heteroskedasticity variance model with outperform counterparts in volatility forecasting in some periods. The third issue concerns the persistent and massive volatility of short-term interest rates. This part inquires how the realizations on Taiwan short-term interest rates can be best described empirically. Various popular volatility specifications are estimated and tested. The empirical findings reveal that the mean reversion is an important characteristic for the Taiwan interest rates, and the level effect exists. Overall, the GARCH-L model fits well to Taiwan interest rates.

Fusion of multiple stereo algorithms is performed in order to obtain accurate disparity segmentation. Reliable disparity map of real-time stereo images is estimated and disparity segmentation is performed for object detection purpose. First, stereo algorithms which have high performance in real-time applications are chosen among the algorithms in the literature and three of them are implemented. Then, the results of these algorithms are fused to gain better performance in disparity estimation. In fusion process, if a pixel has the same disparity value in all algorithms, that disparity value is assigned to the pixel. Other pixels are labelled as unknown disparity. Then, unknown disparity values are estimated by a refinement procedure where neighbourhood disparity information is used. Finally, the resultant disparity map is segmented by using mean shift segmentation. The proposed method is tested in three different stereo data sets and several real stereo pairs. The experimental results indicate an improvement for the stereo analysis performance by the usage of fusion process and refinement procedure. Furthermore, disparity segmentation is realized successfully by using mean shift segmentation for detecting objects at different depth levels.

This study gives our calculations for the temperature-pressure and temperature-concentration phase diagrams using the mean field models applied to ammonium halides (NH4Cl, ND4Cl), ammonium sulfate ((NH4)2SO4/H2O), lithium potassium rubidium sulfate (LiK1-xRbxSO4), potassium pyrosulfate-potassium hydrogensulfate (K2S2O7-KHSO4), cholestanyl myristate-cholesteryl myristate (CnM-CrM), cholestanyl myristate-cholesteryl oleate (CnM-CO), benzene (C6H6) and ice. The phase line equations are derived from the free energies expanded in terms of the order parameters and they are fitted to the experimental data. Some thermodynamic quantities are calculated close to phase transitions in these crystalline systems. We also calculate the specific heat CV using the Raman frequency shifts for NH4Br on the basis of an Ising model close to the lambda-phase transition. A linear relationship is obtained between the specific heat CP and the frequency shifts (1/v)(dv/dT)P near the lambda-point in NH4Br.

The behavior of structures is totally different when they are exposed to fluctuating loading rather than static one which is a well known phenomenon in engineering called fatigue. When the loading is not static but dynamic, the dynamics of the structure should be taken into account since there is a high possibility to excite the resonance frequencies of the structure especially if the loading frequency has a wide bandwidth. In these cases, the structure&rsquo / s response to the loading will not be linear. Therefore, in the analysis of such situations, frequency domain fatigue analysis techniques are used which take the dynamic properties of the structure into consideration. Vibration fatigue method is also fast, functional and easy to implement. In this thesis, vibration fatigue theory is examined. Throughout the research conducted for this study, the ultimate aim is to find solutions to problems arising from test application for the loadings with nonzero mean value bringing a new perspective to mean stress correction techniques. A new method is developed to generate a modified input loading history with a zero mean value which leads in fatigue damage approximately equivalent to damage induced by input loading with a nonzero mean value. A mathematical procedure is proposed to implement mean stress correction to the output stress power spectral density data and a modified input loading power spectral density data is obtained. Furthermore, this method is improved for multiaxial loading applications. A loading history power spectral density set with zero mean but modified alternating stress, which leads in fatigue damage approximately equivalent to the damage caused by the unprocessed loading set with nonzero mean, is extracted taking all stress components into account using full matrixes. The proposed techniques&rsquo / efficiency is discussed throughout several case studies and fatigue tests.

The technique of direct sequence spread spectrum (DS/SS) has been widely used in commercial mobile communication systems. The efficiency of DS/SS system is highly dependent on the accurate and fast synchronization between the incoming and locally generated PN (pseudo-noise) codes. The code synchronization is processed in two steps, acquisition (coarse alignment) and tracking (fine alignment), to bring the delay offset between the two codes. Conventionally, for code synchronization, most of techniques were proposed based on the correlation property of PN codes. Recently, the different approach, by using the adaptive LMS filtering scheme, has been proposed to reduce the hardware complexity and to improve the performance of code synchronization, especially for a long PN code. In this thesis, a new coherent adaptive code synchronization scheme is proposed, where the adaptive constrained LMS (CLMS) algorithm with the maximum tap-weight (MTW) test method is devised for code acquisition. The statistics of weight vector of the proposed CLMS scheme are derived to evaluate the performance, in terms of mean acquisition time (MAT). Analytical and simulation results verify that the proposed scheme for code acquisition outperforms the one using the conventional LMS filtering schemes, under the integer and non-integer time delay cases. Moreover, the setting of threshold value is derived for code acquisition, which is independent of the values of signal-to-noise ratio (SNR) and time delay. Next, the CLMS scheme is proposed associated with the direct delay estimation (DDE) formula for code tracking. This approach does achieve a good delay-tracking performance, which is verified via computer simulation. Simultaneously, the hardware complexity can further be reduced due to that a code-tracking loop implemented by the interpolation method is not required.

The objective of this thesis is to build a phrase recognition system under noisy environment that can be used in real-life. In this system, the noisy speech is first filtered by the enhanced spectral subtraction method to reduce the noise level. Then the MFCC with cepstral mean subtraction is applied to extract the speech features. Finally, hidden Markov model (HMM) is used in the last stage to build the probabilistic model for each phrase. A Mandarin microphone database of 514 company names that are in Taiwan¡¦s stock market is collected. A speaker independent noisy phrase recognition system is then implemented. This system has been tested under various noise environments and different noise strengths.

In this thesis, bit-by-bit power allocation in order to minimize mean-squared error (MSE) distortion of a basic communication system is studied. This communication system consists of a quantizer. There may or may not be a channel encoder and a Binary Phase Shift Keying (BPSK) modulator. In the quantizer, natural binary mapping is made. First, the case where there is no channel coding is considered. In the uncoded case, hard decision decoding is done at the receiver. It is seen that errors that occur in the more significant information bits contribute more to the distortion than less significant bits. For the uncoded case, the optimum power profile for each bit is determined analytically and through computer-based optimization methods like differential evolution. For low signal-to-noise ratio (SNR), the less significant bits are allocated negligible power compared to the more significant bits. For high SNRs, it is seen that the optimum bit-by-bit power allocation gives constant MSE gain in dB over the uniform power allocation. Second, the coded case is considered. Linear block codes like (3,2), (4,3) and (5,4) single parity check codes and (7,4) Hamming codes are used and soft-decision decoding is done at the receiver. Approximate expressions for the MSE are considered in order to find a near-optimum power profile for the coded case. The optimization is done through a computer-based optimization method (differential evolution). For a simple code like (7,4) Hamming code simulations show that up to 3 dB MSE gain can be obtained by changing the power allocation on the information and parity bits. A systematic method to find the power profile for linear block codes is also introduced given the knowledge of input-output weight enumerating function of the code. The information bits have the same power, and parity bits have the same power, and the two power levels can be different.