New range-based neighbourhood operator for extracting edge and texture information from mammograms for subsequent image segmentation and analysis

摘要

A new, spatially isotropic, range-based, neighbourhood operator is described, and its use for extracting edge and texture information from mammograms is illustrated. It is an extension of an operator, first introduced by J.C. Russ, and founded on the work of H.E. Hurst. An octagonal neighbourhood is defined, centred on each pixel in an image, and the difference between the maximum and minimum pixel values in each set of pixels at a given Euclidean distance from the centre pixel is computed to be the range. The logarithm of the range is plotted against the logarithm of the distance, and a straight line fitted to the data. The y-axis intercept c and the square of the correlation coefficient, η~2, are associated with the position of the centre pixel. Preliminary experiments suggest that c is edge-sensitive and could be useful in detecting weak edges, with good noise immunity, while η~2 is a promising texture measure that may be used to detect the edge of the pectoral muscle, define the boundary of the mammographic parenchyma, and in conjunction with other features, possibly detect circumscribed lesions.