Zhiyong Li

Point pattern matching using Relative Shape Context and relaxation labeling

This paper proposes a relative shape context and relaxation labeling (RSC-RL) based approach for point pattern matching (PPM). First of all, a new point set based invariant feature, Relative Shape Context (RSC), is proposed. Using the test statistic of relative shape context descriptors matching scores as the foundation of support function, the point pattern matching probability matrix can be iteratively updated by relaxation labeling (RL). In the end, the one-to-one matching can be achieved by dual-normalization of rows and columns in the finally obtained matching probability matrix. Experiments on both synthetic point sets and real world data show that the performance of the proposed technique is favorable under rigid geometric distortion, noises and outliers.

Applied low dimension linear manifold in hyperspectral imagery anomaly detection

In this paper, a new approach of anomaly detection based on low dimensional manifold will be elaborated. Hyperspectral image data set is considered as a low-dimensional manifold embedded in the high-dimensional spectral space, and this manifold has special geometrical structure, such as Hyper-plane. Usually, the main body of this manifold is constituted by a large area of background spectrum while the anomalistic objects are outside of the manifold. Through the analysis of the geometrical characteristics and the calculation of the appropriate projection direction, anomalistic objects can be separated from background effectively, so as to achieve the purpose of anomaly detection. Experimental results obtained from both the ground and airborne spectrometer data prove effectiveness of the algorithm in improving the detection performance. Since there are no available prior target spectrums to provide proper projected direction, the weak anomalies which have subtle differences from the background on the spectrum will be undetected.

Inexact point pattern matching algorithm based on Relative Shape Context and probabilistic relaxation labelling

The currently known point pattern matching algorithms generally performs poorly when the two point patterns to be matched are not isomorphic. To improve the matching performance of the point pattern matching methods for non-isomorphic point patterns, a novel and robust inexact point pattern matching algorithm that combines with the invariant feature and probabilistic relaxation labelling is proposed. A new point-set based invariant feature, Relative Shape Context (RSC), is proposed firstly. Using the test statistic of relative shape context descriptors matching scores as the foundation of compatibility coefficients, the new support function are constructed based on the compatibility coefficients. Finally, the correct matching results are achieved by using the probabilistic relaxation labelling and imposing the bijective constraints required by the overall correspondence mapping. Experiments on both synthetic point-sets and real image data show that the proposed algorithm is effective and robust.

Anomaly Detection in Hyperspectral Imagery Based on Spectral Gradient and LLE

The local linear embedding algorithm(LLE) is applied into the anomaly detection algorithm on the basis of the feature analysis of the hyperspectral data. Then, to deal with the problem of declining capacity of identifying the neighborhood caused by the Euclidean distance, an improved LLE algorithm is developed. The improved LLE algorithm selects neighborhood pixels according to the spectral gradient, thus making the anomaly detection more robust to the changes of light and terrain. Experimental results prove the feasibility of using LLE algorithm to solve the anomaly detection problem, and the effectiveness of the algorithm in improving the detection performance.

Automatic registration of remote sensing images based on SURF and NSNNI

Image registration is required in many remote sensing applications such as multispectral classification, environmental monitoring, change detection, etc. In this paper, a novel approach of automatic registration of optical remote sensing images based on SURF (Speed Up Robust Features) and NSNNI (Nearest and Second-Nearest Neighbors Iterative Matching) is proposed. Using SURFs detector and descriptor, we can generate scale and rotation invariant control points. Then, the efficient NSNNI method is used to simultaneously find correct matching point pairs and obtain precise transform model. The results of experiments show that our method can achieve sub-pixel accuracy and satisfy the real-time demand.

Hyperspectral subspace estimation preserving anomalies via a test of multivariate sample skewness

Dimensionality Reduction (DR) for hyperspectral image data can be regarded as a problem of signal subspace estimation (SSE) in terms of the Linear Mixing Model (LMM). Most SSE methods for hyperspectral data are based on the analysis of second-order statistics (SOS) without considering preservation of anomalies. This paper addresses the problem of SSE for preserving both abundant and rare signal components in hyperspectral images. The multivariate sample skewness for testing normality is brought in our new algorithm as a discrimination index for rank determination of rare vectors subspace, combining with analysis of the maximum of data-residual ℓ2-norm denoted as ℓ2,∞-norm which is strongly influenced by the anomaly signal components. And the SOS based method, labeled as hyperspectral signal subspace identification by minimum error (HySime), is employed for identification of abundant vectors space. The results of experiments on real AVIRIS data prove that multivariate sample skewness statistics is suitable for measuring the distribution about hyperspectral data globally, and our algorithm can obtain the anomaly components from data that are discarded by HySime, which implies less information loss in the our method.

Improved RX Algorithm with Global Statistics

Anomaly detection of hyperspectral is a hot issue in the remote sensing field. Anomaly detection algorithms currently proposed can be classified into two class, global algorithm and local algorithm. Global algorithm may lead to miss alarm since the discrimination is not accurate enough. On the contrary, local algorithm may bring about false alarm because of lack of global statistics. An improved RX algorithm integrating local and global statistics is proposed. Firstly K-means algorithm is carried out to cluster the whole image into K class which is determined with a virtual dimension estimation method. Then the improved RX is proposed by integrating the global cluster information and the local statistics. Experiment results show that the improved algorithm can obtain a better detection performance than RX algorithm.

Improved ISOMAP algorithm for anomaly detection in hyperspectral images

In this paper, ISOMAP algorithm is applied into anomaly detection on the basis of feature analysis in hyperspectral images. Then an improved ISOMAP algorithm is developed against the limitation existed in ISOMAP algorithm. The improved ISOMAP algorithm selects neighborhood according to spectral angel, thus avoiding the instability of the neighborhood in the high-dimension spectral space. Experimental results show the effectiveness of the algorithm in improving the detection performance.

New Measure Based Manifold Algorithm and Application in Anomaly Detection of Hyperspectral Imagery

Hyperspectral data is endowed with characteristics of intrinsic nonlinear structure and high dimension. In this paper, a nonlinear manifold learning algorithm - ISOMAP is applied to anomaly detection. Then an improved ISOMAP algorithm is developed based on the analysis of the inherent characteristics of hyperspectral imagery. The improved ISOMAP algorithm selects neighborhood according to a novel measure of combination of spectral gradient and spectral angle in order to make the algorithm more robust to the changes of light and terrain. Experimental results prove the effectiveness of the algorithm in improving the detection performance.

Anomaly detection algorithm for hyperspectral images based on background endmember extraction and kernel RX algorithm

The kernel RX algorithm improves the separability between target and background pixels by mapping hyperspectral image data from the low dimensional space into high dimensional feature space. However, the kernel matrix of the background is generated by all image pixels without considering the interference of anomaly target pixels which will make the miss rate increase and consume large memory. To resolve the problem, an anomaly detection algorithm based on background endmember extraction and kernel RX algorithm is introduced. Firstly, the RX algorithm is applied for image processing to filter out obvious anomaly pixels. Then endmember extraction algorithm is used to extract the background endmember according to which the kernel matrix is generated. Experimental results show the effectiveness of the algorithm in improving the detection performance.