Fourier based techniques for the identification of plants and weeds

This paper reviews published data on naturally occurring fractal patterns and their properties, with a view to establishing their use as means for discriminating between plant species. Patterns are called fractal if they look the same at different scales of viewing. Generally, Fourier analysis is shown to be an attractive method of analysing fractals, and can be used to obtain a fractal dimension, which is a characteristic of the pattern shape. Unfortunately, different patterns may have similar such dimensions, but additional methods of separation of such patterns are possible Provided the fractal range is not exceeded, fractal dimensions are shown to be immune to angle of view, or range, thus providing a useful relaxation of image production conditions. The creation of three-dimensional ''terrains'' from two-dimensional images using contrast as a contour level is a possible method of disentangling fractal data contained within an image. Because of natural variations, methods of synthesizing patterns are generally not suitable, unless recognition can be achieved with a limited number of ''characteristic'' basic shapes. Techniques by which shade is converted into shape are inherently unreliable. (C) 1996 Silsoe Research Institute