Abstract

Fractal textures occur in natural scenary or satellite images (clouds, ocean waves). However, processing of fractal image data to extract measurements such as fractal dimension is highly computation-intensive when serial digital methods are used. Massively parallel nonlinear optical processing methods are proposed to characterize fractal images in real time. A photorefractive real time image processor performing a combined convolution/correlation operation is used to measure the correlation dimension of sample fractal patterns. The optical algorithm employs an image autocorrelation performed simultaneously with an optical convolution by using a spatially multiplexed array of 9–100 discs of various radii setting a range of scales. A digital neural network in the correlation plane evaluates the maximum intensity in each light patch and computes the fractal dimension. By placing phase screens or window functions in the third port, additional fractal analyses may be accomplished, including iterated function system encoding. Double convolution involving two different images and a kernel provides scale sensitive image comparison, closely related to wavelet analysis. Real time nonlinear optical coprocessing appears useful in several different approaches to characterization of fractal imagery. Preliminary experimental results will be presented.

© 1992 Optical Society of America