The Detection of Ancient Dwellings Based on Gabor Histogram Features

Abstract

In order to detect objects which may have features of different sizes and directions from satellite map, some bounding boxes which surround these objects are obtained, and convolve with multi-parameters Gabor wavelet to get multi-direction Gabor amplitudes. By computing the average of these Gabor amplitudes, the local texture features of bounding box can obtain, which are direction-invariant. By using histogram algorithm, these Gabor averages are projected onto a given base (some interval divisions) to get different projection coefficients, which can be combined into a feature vector. The vector combines local texture features and global statistic features to representative the texture features of the object. Recognition and Classification of ancient dwellings can be realized by using GLVQ. Experiments show, the vector can representative those irregular and low resolution ancient objects that are based on Gabor direction-invariant local texture features and histogram global features. The GLVQ classifier based the vector of histogram has a good robust, and obtains better detecting effect.