A method of spatial salient structure extraction using local spatial statistics in high resolution images

Homogeneous regions or edges are important structural information for object recognition and extraction in high resolution remote sensing images. This paper considers the homogeneous regions and edges from the perspective of spatial dependence, which is a measure of the spatial association between the pixel values in the image. Spatial dependence is one of the spatial characteristics of high resolution images. Based on the measure to spatial dependence using local spatial statistics (local Moran's I, local Geary's C and Getis), this paper proposes a simple, effective method of extracting spatial salient structures (homogeneous regions or edges) which adopts a new technique of 3D thresholding for spatial dependence intensity. Comparative experiments show the potential and performance differences of three statistics in modeling spatial dependence and extracting spatial salient structures.