A New Built-Up Presence Index Based On Density of Corners

The location and identification of human settlements is a key information in any assessment related to human security and safety decision process, from the preparedness to and mitigation of natural hazards to postdisaster response and reconstruction. The human settlements can be defined as infrastructures build on earth surface for living, trading or any social activity. In order to cover and analyze large areas where scarce and heterogeneous data may exist, the automatic information extraction from Earth Observation is the only sustainable strategy. Several studies about the detection of built-up presence from high resolution satellite were performed during the past few years. Among the various methods, a built-up index aggregating anisotropic textural co-occurence measures (PanTex) was demonstrated to be a robust indicator, where 54 high resolution optical images representing cities distributed around the world are processed. The validation of the built-up presence index emphasized its robustness against the type of constructions, the sun angles and the seasons of acquisitions. In this paper, we propose an alternative procedure for the calculation of built-up presence index from panchromatic high resolution satellite images. While radiometric descriptors of human settlements are highly variable across the world and with illumination conditions, the shape of settlements often contains right angles. This property being more stable, we propose a built-up index based on the density of corners. The index is obtained in two steps: - corners are detected by multi scale Harris detector based on differential morphological decompositions; - corners are spatially aggregated to form a density of corners, which is the built-up index. The differential morphological decomposition is a scale-space representation of the image, where image elements are separated by their scales. Then, the Harris corner indicator, which is highly dependent on a scale parameter, can be adapted to a set of scales. The output of the corner detection is a set of points associated to a scale, which represents the right angles in the image. The corners associated to a scale in the range of admissible settlements dimensions are selected and spatially aggregated to derive a density of corners. The proposed index is extracted from various high resolution panchromatic images and it is compared to the PanTex. In the experimental section, the high correlation between both indicators proves the suitability of the proposed method for consistently detecting built-up presence. Moreover, it gives a new interpretation of the PanTex which is close to a density of corners. Such an observation is critical for understanding the variablity of the PanTex index in between dense urban and industrial areas (industrial areas being composed of large buildings mechanically contain less corners).JRC.G.2-Global security and crisis managemen