Multi-resolution decomposition of fractal images

A method for multi-resolution decomposition of two-dimensional images has been developed in which correlation filters are applied over a range of scales and locally maximum values in the filter outputs extracted. Viewed in the spatial-frequency plane, the correlation filters provide tuned channels, or windows, covering a range of sizes, and can be chosen to act as feature detectors for various feature shapes. Each filter is parameterized by scale L, where L is a distance, and consists of two components: a 'smoothing' component which attenuates fluctuations at scales less than L and a 'differencing' component which extracts two-point differences, or patterns of two-point differences, over distances of order L.<
>