COOCCURRENCE MATRICES FOR IMAGE-ANALYSIS

This paper presents a range of techniques for image segmentation and edge detection based on co-occurrence matrices. Co-occurrence matrices are described and transforms are defined which adapt to global image characteristics and emphasise the differences between typical and atypical image features using co-occurrence matrices as look-up tables. The techniques are extended by analysing the matrices and labelling them, with the result that a labelled matrix can be used to segment the regions of an image and to simultaneously detect prominent edges. Examples are given for a variety of images: synthetic, infra-red, multispectral and temporal. It is also shown that the techniques presented can be integrated easily with a variety of postprocessing techniques such as hysteresis and relaxation labelling for enhanced performance.