Localisation of image features using measures of rank distribution

Measures which characterise local rank distribution can be used to enhance and localise image structure. The diversity and local rank of pixels provide results which are invariant to any strictly order-preserving transformations previously applied to the image data. Localisation of image features to pixel precision is possible through the rank distribution, even with the use of relatively large window sizes. To help detect variations in rank distributions over lengths smaller than the window size, we present methods to sharpen the response of local diversity to image structure. The aim is to enhance image based directly on rank, rather than using conventional linear edge detection. The methods outlined in this paper may be applied to single or multi-band image data