Residential area detection from high-resolution remote sensing imagery using corner distribution

Traditional residential area detection methods are mainly based on image features, such as texture, spectrum, shape and etc. However, these features are not invariant to scale and illumination changes, which consequently reduce the robust of the existing algorithms. To solve this problem, the proposed method uses local feature for residential area detection from high-resolution remote-sensing imagery, which consists of three steps. Firstly, a large set of local feature points are extracted by Harris corner detector. In order to achieve a reliable extraction of corners from residential areas, two criterions are further proposed to validate and filter them. Afterwards, the extracted corners are incorporated into a likelihood function, and are used to measure the possibility of each pixel belonging to the residential area. Finally, residential areas are extracted by an adaptive binary segmentation method. Experimental results show that the proposed approach outperforms the existing algorithms in terms of detection accuracy.