Abstract: House goals are separated from satellite image by using rectangle algorithm based on area feature and shape feature, and the mistakenly identified objects need to be eliminated from all separated objects based on their different texture features. Rotating, enlarging and cutting operations need to be designed to make those separated objects standardize with same size and direction. Elastic grid technique can choose several characteristic lines and columns of the image, and the intersections of them can partition an image into several characteristic grids. The classic relevant statistics of gray level co-occurrence matrix (GLCM) of each grid are computed to generate a characteristic fingerprint array, which comprehensively reflect the local texture features of the grid. The combination of fingerprint arrays of all grids can generate the fingerprint vector of an image, which can reflect global features of the image. The feature vector generated from improved elastic grid partition and GLCM statistic can simultaneously characterize the local texture features and global statistic features of an image. Ancient houses are accurately identified and classified by the similarity comparison to house samples of different periods based on their characteristic fingerprint vectors. Experiments show that the correct rate of separated house targets is about 86.9% by using rectangle algorithm, and using the house fingerprint algorithm based on elastic grid partition and GLCM statistics, the correct preliminary classification rate of ancient houses is more than 97.4%.

Key words: rectangle algorithm, gray level co-occurrence matrix (GLCM), elastic grid, fingerprint, house classification

摘要： 使用基于面积特征和形状特征的矩形算法可以从卫星图像中分离出民居目标，对于误识别疑似民居区域，需要进一步提取它们的纹理特征加以排除。并且需要设计旋转、放大和裁剪算子，对所有抽取目标进行大小和方向的标准化。弹性网格技术可以选取图像的多个特征行和特征列，而它们的相交把一个图像划分成多个特征子格。计算出每个子格的灰度共生矩阵（gray level co-occurrence matrix，GLCM）的几个经典特征值形成一个特征数组，可以反映子格的局部纹理特征。所有子格的特征数组顺序组合形成一个特征向量，可以反映这个图像的全局特征。基于改进弹性网格划分和子格GLCM特征值的指纹向量能够同时表征一个图像的局部纹理特征和全局统计特征。通过与不同年代的民居样本特征指纹的相似度比较，实现了古民居的精确识别与分类。实验表明，使用矩形算法抽取出民居目标的正确率为86.9%，使用基于弹性网格划分和GLCM特征值的民居指纹算法，古民居初步分类正确率超过97.4%。

关键词: 矩形算法, 灰度共生矩阵（GLCM）, 弹性网格, 指纹, 民居分类