Building extraction in satellite images using active contours and colour features

ABSTRACT Obtaining the segmentation of building footprints from satellite images is a complex process since building areas and their surroundings are presented with various colour intensity values and complex features. Active contour region-based segmentation methods can be used to establish the corresponding boundary of building structures. Typically, these methods divide the image into regions that exhibit a certain similarity and homogeneity. However, using the traditional active contour algorithms for building structures detection, in several cases where spectral heterogeneity exists, over-detection or under-detection are usually noticed. In this work, the Red, Green and Blue (RGB) representation and the properties of the Hue, Saturation and Value (HSV) colour space have been analysed and used to optimize the extraction of buildings from satellite images in an active contour segmentation framework. Initially, the satellite image was processed by applying a clustering technique using colour features to eliminate vegetation areas and shadows that may adversely affect the performance of the algorithm. Subsequently, the HSV representation of the image was used and a new active contour model was developed and applied for building extraction, utilizing descriptors derived from the value and saturation images. A new energy term is encoded for biasing the contours to achieve better segmentation results. An effective procedure has been designed and incorporated in the proposed model for the active contour initialization. This process enhances the performance of the model, leading to lower computational cost and higher building detection accuracy. Additionally, statistical measures are used for designing optimum morphological filters to eliminate any misleading information that may still exist. Qualitative and quantitative measures are used for evaluating the performance of the proposed method.