A new approach is proposed to extract the rectangular building footprints from single high-resolution synthetic aperture radar (SAR) image by combining region-based information with edge-based information. We build multiscale autoregressive (MAR) model and likelihood rate classifier for SAR image to extract region information. Edge information is extracted from SAR image by using constant false alarm rate edge detector. We propose a new optimization criterion which fuses the previous extracted region information and edge information to obtain the final building footprints. The experimental result shows that the proposed method can extract the building footprints effectively.