基于三维纹理分割的高光谱图像异常检测

一幅复杂背景的高光谱图像可以看成是由不同纹理组合而成,纹理的统计特性可以近似用高斯分布来描述。采用纹理分割实现复杂背景的分解,从而突破异常大小和形状的限制。采用三维高斯马尔科夫场来描述高光谱图像背景的分布特性,利用最大似然估计得出模型参数,以此参数为特征进行纹理分割,在各纹理上计算像素的统计特性,进行异常检测。

Anomaly detection based on 3 D texture segmentation for hyperspectral image

A hyperspectral image with complex background can be regarded as the components of different textures,the statistical distribution characteristic of which can be described by Gaussian model.Based on the model,a new algorithm is designed to segment the complex background into homogenous regions,breaking through the limits of size and form of the anomaly regions.The background distribution characteristic of hyperspectral image is described by 3 D Markov model.The model parameters are obtained by the Max Likelyhood Estimation algorithm.Then the texture segmentation and the anomaly detection algorithm are achieved based on the model parameters and the pixel statistical characteristic respectively.