Unsupervised classifying land development in high resolution multi-spectral satellite photos and application for facial features extraction

A novel method of unsupervised detection of the different types of land development using multi-spectral IKONOS satellite photographs is proposed. The method is based on image decomposition into two clusters: the type of land development in quest and the rest of the image, using optimal (the most informative) combination of color and spatial homogeneity indicators. Experimental study of the proposed technique shows fast and reliable deciphering of the vegetation, bare soils, water, roads and residential areas. The results of this preliminary classification are used for accurate decipherment of terrain details using high resolution BW IKONOS photography, which comes along with the multi-spectral one. It is shown that proposed technique is suitable for facial features extraction as well. Fast and reliable detection of human skin, eyes, mouth, hair face triangle and oval is obtained on different subjects.