Ke Youan

Human face orientation estimation using symmetry and feature points analysis

A new method for human face orientation estimation based on continuous symmetry and feature points analysis is presented using Cn and Dn continuous symmetry measuring. The quantitative measure of face orientation can be calculated. In this paper a group of still images and a video sequence are processed using the above method. The experiment shows that the presented orientation estimation method works well and could effectively reflect the range of human face orientations and the calculation results are robust to the scale, position and angles of the objects.

Surface extraction for volume rendering based on zero crossing points detection in wavelet domain

Surface extraction is a popular preprocessing method for volume rendering acceleration. We discuss a surface extraction algorithm based on the wavelet transform. Given a basis wavelet function meeting certain conditions, the points on object surfaces can be picked up by detection of zero crossing points from the 3D wavelet transform components of the volume data. The experimental results show that multi-scale surfaces of objects can be obtained for further display. With the reduction of the effective samples for rendering, the speed of display increases with the tolerable cost of image quality.

A genetic algorithm with sharing and its application in MUSIC method

Many optimization problems require the investigation of multiple local extremities and the original genetic algorithms with sharing cannot do this with objective functions with non-uniformly distributed extremities. In this paper, a new sharing method is developed. We use the estimation of the directions of spatial emitters as an example to test the power of our new method. The objective functions are the spectra estimated by the MUSIC method. It is shown by simulation that the new algorithm can find all the local extremities of the objective functions and its computation consumption is at most one-tenth of the conventional MUSIC method.

Weighted statistical model for non-rigid image objects detection

A new weighted statistical model is presented for detecting non-rigid image targets. In the detection or matching stage, a modified GA (genetic algorithm) is presented to solve the problem of adaptability to scaling, position and tilted angles. We demonstrate our WSM model on the practical human face detection problem. Some results are presented.

Hybrid natural mode and scattering center method for the representation of radar target impulse response

The singularity expansion method which has received a great deal of attention over the last two decade years is an effective computational tool in dealing with the impulse response of a radar target. However, the incompleteness of the singularity expansion method involving the resonance frequencies of a radar target is now well known and well understood. Indeed, in the early time, the complex resonance model is an improper model and must be supplanted, or supplemented, by techniques that consider the local features of the scattering object. A new hybrid method is considered by using the singularity expansion and scattering center methods involving the singularity of the surface of a radar target. This method should make it better to illuminate the scattering mechanism of a radar target.

Study on radar range profiles for target recognition

In the high frequency region, the radar range profiles are chosen as feature detectors for target recognition. The profile is dependent on the aspect of the target. In order to make radar target recognition (RTR) more effective, the definitions of correlation length and correlation angle are given. In the frequency domain, the most important feature of the shape of a profile is mainly contained in its spectral phase not in its spectral magnitude. So the spectral phase holds important feature information for RTR in the high frequency region, and should be utilized effectively.

Automatic target recognition based on correlation matrix memory

Correlation matrix memory is one of the models of the association memory, and can be applied to high range resolution radar target recognition. In this paper, the association memory matrix is used to memorize a great number of encoded profiles belong to different targets. Results show that it is robust to noise and aspect changes.

A two-step adaptive beamforming approach

A two-step adaptive beamforming approach is introduced, which forms the desired radiation pattern through linear combination of a family of auxiliary beams after the number and locations of spatial emitters are obtained by spectral estimation. This algorithm has excellent performance in jamming cancellation, stability, and computational complexity.

A fast fractal image coding scheme

A new fast fractal image coding (FFIC) scheme is presented, in which the fractal parameters are directly obtained from the range and domain blocks. Compared with the conventional fractal image coding (FIC) scheme, FFIC is six times faster in compression speed with a small loss of PSNR of about 0.5 dB.

Multiple polyphase signals for access system

We present several theorems which are very attractive in designing multiple signals for an access system. The designed signals are all derived from the orthogonality of the Frank code matrix, and they are associated with complementary codes or perfect periodic codes. The co-interference between any two signals in a certain set can be reduced to satisfy the needs of a multiple access system. We also discuss the Doppler property of the designing codes and propose an architecture for their implementation.<<ETX>>