A new range-based image texture measure with application to mammogram image analysis

This paper describes a new spatially isotropic, neighbourhood-based, image texture measure, and illustrates its use for mammographic image analysis. It is based on the extended Russ operator, first introduced by J.C. Russ (1990). 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 d from the centre pixel is computed to be the range d. Log d is plotted against log d, and a straight line fitted to the data. The square of the correlation coefficient, /spl eta//sup 2/, is associated with the position of the centre pixel. Preliminary experiments suggest that /spl eta//sup 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.