This thesis outlines the development of four image processing algorithms that combine spectral and spatial information for the detection of complex objects on the Earth' surface by remote sensing. Complex objects are objects that are composed of several smaller parts. These smaller parts may not be spectrally unique in themselves and lack a statistical coherence, thus creating a problem for image processing techniques that operate per pixel. As the human mind is often capable of processing and recognizing such objects by combining spectral and spatial information, a solution to this problem is an image processing technique that is based on the same qualitative reasoning. The thematic focus is on the detection of natural hydrocarbon seepages. Natural hydrocarbon seepages are non-unique in both their spectral and spatial characteristics. The infrared spectrum of oil is easily confused with other bituminous surfaces such as asphalt. The dominant anomaly that results from the presence of hydrocarbons is a circular halo of bare soil. The first algorithm measured the shape of a homogeneous object, based on relations between the area and perimeter of an object and its convex hull. The second algorithm was a template matching algorithm that matched a miniature image to a remotely sensed image. Both algorithms were based on existing techniques and were shown to perform well in combining spectral and spatial information. It is however concluded that both technigues were not enough versatile for the detection of seepage-induced halos. The third and fourth algorithms were specifically designed for the detection of seepage-induced halos. The third algorithm aimed to detect spectrally homogeneous pixels on a circle. The fourth algorithm was an extension of the third, aimed to detect incomplete circles with variable radii and was based on Hough transforms. Results showed that these algorithms can detect halos of bare soil that result from seeping hydrocarbons. It was finally concluded that the knowledge-based spatial-spectral approach is an improvement of over traditional remote sensing image processing methods in the detection of anomalies resulting from natural hydrocarbon seepages. The concept of the presented algorithms in general and the design of these algorithms in particular allow them to be applied to other remote sensing research.
