| Abstract |
For decades, fractals have been put forward as the de-facto model for many natural and social complex phenomena: from the shape of a coast line, to the price fluctuations in the stock market. Given a fractal model it is a simple exercise to write an algorithm that draws it, or samples data from it. The reverse is not so simple: given a fractal image, or fractal data, how do we find the model that produced it? This is called the fractal inverse problem and the lack of workable solutions has been one of the main problems holding the field back. We present a new approach based on the classic Expectation-Maximization algorithm. |