Li Jingwen

Facial expression recognition based on two dimensional feature extraction

In this paper, approaches to facial expression recognition based on two dimension methods are studied. 2D expression feature extraction methods include 2DPCA, 2DLDA and generalized low rank approximation of matrices (GLRAM) which are usually used in face recognition or data compression. After expression features are extracted, support vector machine classifier is used for expression classification. Extensive expression recognition experiments are carried out on Japanese female facial expression database (JAFFE) to study the influence of feature dimension on recognition ratio and the results are also compared with that of the tradition 1D feature extraction methods such as PCA and LDA. Experiment results show that 2D methods are effective in expression recognition application and usually outperform 1D methods.

Analysis on noise reduction method for interferometric SAR image

Noise in SAR interferogram brings much trouble in phase unwrapping. It will also influence the accuracy of digital elevation models. Based on realization of complex mean filtering, pivoting mean filtering, pivoting median filtering three noise reduction algorithms, their computations are analyzed and the effect of noise reduction is given out. In order to get the most optimized balance between computation and effect of noise reduction, a new noise reduction idea - second-time or multiple combined noise reduction is proposed. The algorithm can comprehend the merits of several noise reduction methods. It can preserve edge information well and decrease phase noise dramatically at the cost of adding a little computation. Based on computer simulation, certain terrain is set and the effect of noise reduction of new method is verified. The variance of before and after combined noise reduction is given out. The effect of three-time or more combined noise reduction is not obvious because noise has been reduced by second-time noise reduction. We do not recommend 4 or more-time combined noise reduction for its large computation. ERS-frac12 real data are also used to verify the validity of the new algorithm

Facial Expression Recognition Based on Classifier Combinations

This paper introduces facial expression recognition by combining multiple classifiers. It describes in detail the principles of various classifier combination schemes. Extensive experiments of these classifier combination schemes on both the Japanese female facial expression database and the Cohn-Kanade AU-coded facial expression database are given. The results show that it is an effective approach to further increase the recognition rate by classifier combinations

Motion error analysis and compensation for airborne formation flying InSAR

Because the stability of plane platform is not good, airborne SAR can not acquire the precise orbit parallelism and velocity as spaceborne SAR. Motion error problem for airborne formation flying InSAR is analyzed here. The plane interval changes (across track baseline changes) can be discussed in three situations: interval increasing, interval decreasing, interval changing according to sine curve. Phase changes with baseline changes are deduced quantiticationally. The simulation result of plane interval changes is given out. It can be concluded that across track baseline changes lead to azimuth slope. The digital relation between across-track baseline error and phase error, height error is concluded by simulation experiment result. The paper presents the method of azimuth slope elimination. Azimuth slope eliminating is adopted in simulation terrain. Two control points are set in topographic area to compensate the phase error of baseline changes. This algorithm can improve the DEM precision for whole area. Even if the plane track is curve, the method in this paper can still compensate most of the motion error.

An algorithm for interferometric SAR data processing

In this paper, an algorithm of generating INSAR unwrapped phase image from SAR single-look complex images is presented. Besides the general processing technique, this article focuses on the methods of flat-earth phase removal, phase noise reduction and phase unwrapping. The availability is tested by the results of processing ERS-1/2 SAR images.

An airborne squint SAR processing algorithm

An airborne squint synthetic aperture radar (SSAR) imaging processing method using a chirp scaling (CS) algorithm is presented. The cubic phase in the azimuth spectrum is considered in the chirp scaling phase factor and azimuth factor. Compared with the range/Doppler (RD) algorithm and wavenumber algorithm, the algorithm not only completes range cell migration correction (RCMC) precisely, but also avoids interpolation, and requires only complex multiplies and fast Fourier transforms (FFTs) to implement the processing. The CS algorithm provides an image quality that is better than the precision RD algorithm. This paper gives computer simulation results of the radar imaging. The results show that this imaging method is effective.

Cluster-Based Split-Window Radon Transform Algorithm for Ship Wake Detection

The purpose of this article is to present a novel algorithm for ship wake detection in synthetic aperture radar (SAR) images. The main originality of our work is that splitting the image with small window before conventional Radon transform to make the illumination has stronger consistency in each window and adopting clustering algorithm to select real wakes form disturbing lines. Experimental result on real SAR image is presented and compared to that obtained using conventional approaches.

An improved ATI method on airborne SAR ground slow moving target detection

Based on airborne dual channel SAR, this paper improved AT I technique, put forward a weighting factor from DPCA to AT I to improve ground slow velocity moving target detect capability, the method integrated the advantage of DPCA and ATI technique, using the weight value received from DPCA to weighting process the result of interferential ATI, further more, bringing forward a new moving target detect method. The outcome of computer emulation indicates that compared with ATI method, this papers method can effectively decrease false target number and detect feebleness target efficiency. Compared with DPCA method, it can detect slower moving targets which DPCA cannot detect because of excessive counteraction.

Computation of Optimum Azimuth Sampling Time for Dual-Channel DPCA System based on Signal-to-Noise Ratio

This paper starts on the flow of moving target detection in dual channel DPCA system. Firstly the output power density spectrum of moving target and noise in DPCA system is analyzed, and the output SNR is obtained. Then the effect of synthetic aperture time of moving targets echo and azimuth antenna pattern on output SNR in DPCA is discussed in detail. Based on this effect, the paper deduces the expression which describes the variation of output SNR against synthetic aperture time in Dual-Channel DPCA system, where azimuth pattern is Sinc function, and the synthetic aperture time for optimizing the ability of detecting the moving target is obtained. Further more, this method of acquiring the aperture time can be applied for other antenna patterns. It is testified by processed result of simulated data that the relationship between output SNR and synthetic aperture time fits the above conclusion in Dual-Channel DPCA system.

Precise Simulation of Spaceborne Synthetic Aperture Radar and Its Evaluation

Simulation technique becomes increasingly important with the development of Synthetic Aperture Radar (SAR). It can be applied for many purposes such as testing different imaging algorithms and validating different SAR system design parameters[1]. For improving the precision of simulator determination of the location (mainly refers to latitude) of satellite which illuminates the fixed scene center is reconsidered in this paper. As attitude of satellite strongly affects Doppler parameters, determination of attitude is also discussed for simulated system approaching real system. Although attitude determination based on Doppler centroid was studied in many articles[2][3], this paper utilizes the information about the footprint shape of beam, and develops the equations relating yaw, pitch and especially roll to slant range and Doppler shift of echo data. We can get the attitude of satellite by solving these equations. Then for validating and evaluating the simulated system, this paper describes a novel approach which involves imaging combination of real and simulated echo data. Doppler parameters adopted in imaging algorithm before and after combining are identical. Effect of focalizing simulated echo data indicates whether the simulated system is acceptable, and relative error of geometric distortion which appears in the SAR image reflects the approaching degree of simulated system to real system. Test was conducted by generating simulated data for RADARSAT-1, the result shows the precision of location and attitude determination and validation of this evaluation approach.