Rui Yao

Faint moving object detection in optical astronomical image using a 2d-1d IUWT

In this work, we introduce a method for the detection of faint moving object in optical astronomical image sequence. It is based on multiscale 2D-1D image analysis resulting from the a trous wavelet transform decomposition of the original image sequence. The sequence is decomposed into 2D-1D spatial-temporal IUWT(Isotropic Undecimated Wavelet Transform) coefficients. We get multiscale product of 2D-1D wavelet coefficients after thresholding of non-significant coefficients. The multiscale correlation of the filtered wavelet coefficients combines multiscale product coming from different levels of resolution and gives a distinctive characterization of the moving object. By accumulating the multiscale product of each image, the intensity of moving objects will be increased, which increases the accuracy of localization. Experiments performed on optical astronomical image data demonstrate the robustness and efficiency of the method.

Generalized kernel density estimation based robust estimator and its application

In this paper, a new General Kernel Density Estimator (GKDE) based robust estimator is presented. The GKDE based robust estimator utilizes GKDE to estimate the distribution of data points and by using local adaptive bandwidth estimator, the scale of inliers or user-specified error threshold is not need. Compared to ASKC, pbM and other Kernel Density Estimation based robust estimator which do not have locality, GKDE has higher resolution for inliers, and experiments show that it has higher precision than traditional robust estimator such as RANSAC, LMeds. We also applied GKDE based estimator to image mosaic for homography estimation.

Out-of-Sample embedding of spherical manifold based on constrained least squares

All of the current state-of-the-art nonlinear dimensionality reduction methods attempt to seek the low-dimensional manifold structure by preserving global or local properties of the original data, but without considering the constraint of the manifold structure, thus, there may be a big contrast between the manifold structure result obtained by the nonlinear techniques and the result that we expected. Therefore, it is necessary for us to study the constrained nonlinear dimensionality reduction. In this paper, we study the embedding of a hidden manifold onto a unit sphere by using SMACOF algorithm and propose a method to solve the out-of-sample problem which usually arises in the manifold learning. By converting it into a constrained least squares problem with the spherical structure information, this method avoids reconstructing the neighborhood graph. The application results of 3-D object pose estimation show the effectiveness of our propose method.

A robust smear removal method for inter-frame Charge-Coupled Device star images

In order to eliminate the influence of Smear Effect on follow-up processing of star images, this paper researched the source and statistical model of Smear Effect. After researching the working progress of inter-line Charge Coupled Device(CCD), inter-frame CCD and full-frame CCD, this paper builds a statistical model for the background noise and then proposes an algorithm to do radiometric correction in smear images based on modeling and estimating the intensity of background noise in star image. Experimental results indicate that the algorithm in this paper can remove smear effect in star image efficiently while retaining origin information. The method in this paper can eliminate the influence of smear effect in star images while retaining origin information.

Small space target detection based on feature stability algorithm

The detecting technology is very important for the discovery of space objects. The target is submerged by the complex background noise which is the environment of outer space and the device produced. The difficulty of the detection is increased. The detect system mainly has three kinds of working patterns including the fixed tracking pattern, the fixed star tracking pattern and the target tracking pattern. Theses pattern differences perform as moving target in moving background, moving target in static background and static target in moving background in the images. We bring up a new framework for detecting target in three kinds of working patterns based on the feature stability difference. The first step is preprocess. Secondly, we extract features from image sequence. Then we construct a stability function about features for every pixel. Finally, we can detect the position of target according to the value of stability function, then map the position of target in the feature domain to the original image, and search in original image for the accurate centroid of the target. Qualitative and quantitative results prove that the proposed algorithm has strong anti-noise performance and fit for kinds of working pattern of detection system for target detection conveniently.

Smear radiometric correction algorithm in star images based on kernel density estimation

In order to eliminate the influence of Smear Effect on follow-up processing of star images, this paper researched the source and statistical model of Smear Effect. After researching the working progress of inter-line Charge Coupled Device(CCD), inter-frame CCD and full-frame CCD, this paper builds a statistical model based on kernel density estimation for the background noise and then proposes an algorithm to do radiometric correction in smear images based on modeling and estimating the probability density function of background noise in star image. Experimental results indicate that the algorithm in this paper can remove smear effect in star image efficiently while retaining origin information. The method in this paper can eliminate the influence of smear effect in star images while retaining origin information.